app.py

import urllib.parse

import gradio as gr
from starlette.middleware.base import BaseHTTPMiddleware
from starlette.responses import Response
from starlette.types import ASGIApp
from fastapi import status, FastAPI

class BlockFileRedirectMiddleware(BaseHTTPMiddleware):
    def __init__(self, app: ASGIApp):
        super().__init__(app)

    async def dispatch(self, request, call_next):
        # URL decode the path first
        path = urllib.parse.unquote(request.url.path)

        # Check all possible file endpoint patterns
        if any(
            pattern in path
            for pattern in [
                "/gradio_api/file=",
                "/gradio/api/file=",
                "/api/file=",
                "/file=",
            ]
        ):
            # Extract everything after file=
            file_part = path.split("file=", 1)[1]
            # Block if it's a URL (starts with http:// or https://)
            if file_part.lower().startswith(("http://", "https://")):
                return Response(
                    status_code=status.HTTP_403_FORBIDDEN,
                    content="Direct URL redirects are not allowed",
                )
        return await call_next(request)

import pandas as pd
import requests
from docx import Document
import os
from openai import OpenAI
from groq import Groq
import uuid
from gtts import gTTS
import math
from pydub import AudioSegment


from youtube_transcript_api import YouTubeTranscriptApi
from youtube_transcript_api._errors import NoTranscriptFound
import yt_dlp

from moviepy.editor import VideoFileClip
from pytube import YouTube
import os
import io
import time
import json
from datetime import datetime, timezone, timedelta
from urllib.parse import urlparse, parse_qs

from google.cloud import storage
from google.cloud import bigquery
from google.oauth2 import service_account
from googleapiclient.discovery import build
from googleapiclient.http import MediaFileUpload
from googleapiclient.http import MediaIoBaseDownload
from googleapiclient.http import MediaIoBaseUpload

from educational_material import EducationalMaterial
from storage_service import GoogleCloudStorage
from google.oauth2.service_account import Credentials
import vertexai
from vertexai.generative_models import GenerativeModel, Part

# import boto3

from chatbot import Chatbot

is_env_local = os.getenv("IS_ENV_LOCAL", "false") == "true"
print(f"is_env_local: {is_env_local}")

print("===gr__version__===")
print(gr.__version__)

# KEY CONFIG
if is_env_local:
    with open("local_config.json") as f:
        config = json.load(f)
        IS_ENV_PROD = "False"
        PASSWORD = config["PASSWORD"] 
        GCS_KEY = json.dumps(config["GOOGLE_APPLICATION_CREDENTIALS_JSON"])
        DRIVE_KEY = json.dumps(config["GOOGLE_APPLICATION_CREDENTIALS_JSON"])
        GBQ_KEY = json.dumps(config["GOOGLE_APPLICATION_CREDENTIALS_JSON"])
        OPEN_AI_KEY = config["OPEN_AI_KEY"]
        OPEN_AI_ASSISTANT_ID_GPT4_BOT1 = config["OPEN_AI_ASSISTANT_ID_GPT4_BOT1"]
        OPEN_AI_ASSISTANT_ID_GPT3_BOT1 = config["OPEN_AI_ASSISTANT_ID_GPT3_BOT1"]
        GROQ_API_KEY = config["GROQ_API_KEY"]
        JUTOR_CHAT_KEY = config["JUTOR_CHAT_KEY"]
        AWS_ACCESS_KEY = config["AWS_ACCESS_KEY"]
        AWS_SECRET_KEY = config["AWS_SECRET_KEY"]
        AWS_REGION_NAME = config["AWS_REGION_NAME"]
        OUTPUT_PATH = config["OUTPUT_PATH"]

else:
    IS_ENV_PROD = os.getenv("IS_ENV_PROD", "False")
    PASSWORD = os.getenv("PASSWORD")
    GCS_KEY = os.getenv("GOOGLE_APPLICATION_CREDENTIALS_JSON")
    DRIVE_KEY = os.getenv("GOOGLE_APPLICATION_CREDENTIALS_JSON")
    GBQ_KEY = os.getenv("GOOGLE_APPLICATION_CREDENTIALS_JSON")
    OPEN_AI_KEY = os.getenv("OPEN_AI_KEY")
    OPEN_AI_ASSISTANT_ID_GPT4_BOT1 = os.getenv("OPEN_AI_ASSISTANT_ID_GPT4_BOT1")
    OPEN_AI_ASSISTANT_ID_GPT3_BOT1 = os.getenv("OPEN_AI_ASSISTANT_ID_GPT3_BOT1")
    GROQ_API_KEY = os.getenv("GROQ_API_KEY")
    JUTOR_CHAT_KEY = os.getenv("JUTOR_CHAT_KEY")
    AWS_ACCESS_KEY = os.getenv("AWS_ACCESS_KEY")
    AWS_SECRET_KEY = os.getenv("AWS_SECRET_KEY")
    AWS_REGION_NAME = 'us-west-2'
    OUTPUT_PATH = 'videos'

TRANSCRIPTS = []
CURRENT_INDEX = 0
CHAT_LIMIT = 5

# Google aiplatform
google_service_account_info_dict = json.loads(GBQ_KEY)
GOOGPE_SCOPES = ["https://www.googleapis.com/auth/cloud-platform"]
google_creds = Credentials.from_service_account_info(
    google_service_account_info_dict, scopes=GOOGPE_SCOPES
)
vertexai.init(
    project="junyiacademy",
    service_account=google_service_account_info_dict,
    credentials=google_creds,
)

# CLIENTS CONFIG
GBQ_CLIENT = bigquery.Client.from_service_account_info(json.loads(GBQ_KEY))
GROQ_CLIENT = Groq(api_key=GROQ_API_KEY)
GCS_SERVICE = GoogleCloudStorage(GCS_KEY)
GCS_CLIENT = GCS_SERVICE.client
# BEDROCK_CLIENT = boto3.client(
#     service_name="bedrock-runtime",
#     aws_access_key_id=AWS_ACCESS_KEY,
#     aws_secret_access_key=AWS_SECRET_KEY,
#     region_name=AWS_REGION_NAME,
# )

# check open ai access
def check_open_ai_access(open_ai_api_key):
    # set key in OpenAI client and run to check status, if it is work, return True
    client = OpenAI(api_key=open_ai_api_key)
    try:
        response = client.chat.completions.create(
            model="gpt-3.5-turbo",
            messages=[
                {"role": "user", "content": "This is a test."},
            ],
        )
        if response.choices[0].message.content:
            return True
        else:
            return False
    except Exception as e:
        print(f"Error: {str(e)}")
        return False

open_ai_api_key_assistant_id_list = [
    {
        "account":"bot1", 
        "open_ai_api_key": OPEN_AI_KEY, 
        "assistant_gpt4_id": OPEN_AI_ASSISTANT_ID_GPT4_BOT1, 
        "assistant_gpt3_id": OPEN_AI_ASSISTANT_ID_GPT3_BOT1,
    },
]
for open_ai_api_key_assistant_id in open_ai_api_key_assistant_id_list:
    account = open_ai_api_key_assistant_id["account"]
    open_ai_api_key = open_ai_api_key_assistant_id["open_ai_api_key"]
    if check_open_ai_access(open_ai_api_key):
        OPEN_AI_CLIENT = OpenAI(api_key=open_ai_api_key)
        OPEN_AI_ASSISTANT_ID_GPT4 = open_ai_api_key_assistant_id["assistant_gpt4_id"]
        OPEN_AI_ASSISTANT_ID_GPT3 = open_ai_api_key_assistant_id["assistant_gpt3_id"]
        print(f"OpenAI access is OK, account: {account}")
        break

# 驗證 password
def verify_password(password):
    if password == PASSWORD:
        return True
    else:
        raise gr.Error("密碼錯誤")

# # ====drive====初始化
def init_drive_service():
    credentials_json_string = DRIVE_KEY
    credentials_dict = json.loads(credentials_json_string)
    SCOPES = ['https://www.googleapis.com/auth/drive']
    credentials = service_account.Credentials.from_service_account_info(
            credentials_dict, scopes=SCOPES)
    service = build('drive', 'v3', credentials=credentials)
    return service

def create_folder_if_not_exists(service, folder_name, parent_id):
    print("检查是否存在特定名称的文件夹，如果不存在则创建")
    query = f"mimeType='application/vnd.google-apps.folder' and name='{folder_name}' and '{parent_id}' in parents and trashed=false"
    response = service.files().list(q=query, spaces='drive', fields="files(id, name)").execute()
    folders = response.get('files', [])
    if not folders:
        # 文件夹不存在，创建新文件夹
        file_metadata = {
            'name': folder_name,
            'mimeType': 'application/vnd.google-apps.folder',
            'parents': [parent_id]
        }
        folder = service.files().create(body=file_metadata, fields='id').execute()
        return folder.get('id')
    else:
        # 文件夹已存在
        return folders[0]['id']

# 检查Google Drive上是否存在文件
def check_file_exists(service, folder_name, file_name):
    query = f"name = '{file_name}' and '{folder_name}' in parents and trashed = false"
    response = service.files().list(q=query).execute()
    files = response.get('files', [])
    return len(files) > 0, files[0]['id'] if files else None

def upload_content_directly(service, file_name, folder_id, content):
    """
    直接将内容上传到Google Drive中的新文件。
    """
    if not file_name:
        raise ValueError("文件名不能为空")
    if not folder_id:
        raise ValueError("文件夹ID不能为空")
    if content is None:  # 允许空字符串上传，但不允许None
        raise ValueError("内容不能为空")
    
    file_metadata = {'name': file_name, 'parents': [folder_id]}
    # 使用io.BytesIO为文本内容创建一个内存中的文件对象
    try:
        with io.BytesIO(content.encode('utf-8')) as fh:
            media = MediaIoBaseUpload(fh, mimetype='text/plain', resumable=True)
            
            print("==content==")
            print(content)
            print("==content==")

            print("==media==")
            print(media)
            print("==media==")
            # 执行上传
            file = service.files().create(body=file_metadata, media_body=media, fields='id').execute()
            return file.get('id')
    except Exception as e:
        print(f"上传文件时发生错误: {e}")
        raise  # 重新抛出异常，调用者可以根据需要处理或忽略

def upload_file_directly(service, file_name, folder_id, file_path):
    # 上傳 .json to Google Drive
    file_metadata = {'name': file_name, 'parents': [folder_id]}
    media = MediaFileUpload(file_path, mimetype='application/json')
    file = service.files().create(body=file_metadata, media_body=media, fields='id').execute()
    
    
    # return file.get('id')  # 返回文件ID
    return True

def upload_img_directly(service, file_name, folder_id, file_path):
    file_metadata = {'name': file_name, 'parents': [folder_id]}
    media = MediaFileUpload(file_path, mimetype='image/jpeg')
    file = service.files().create(body=file_metadata, media_body=media, fields='id').execute()
    return file.get('id')  # 返回文件ID

def download_file_as_string(service, file_id):
    """
    从Google Drive下载文件并将其作为字符串返回。
    """
    request = service.files().get_media(fileId=file_id)
    fh = io.BytesIO()
    downloader = MediaIoBaseDownload(fh, request)
    done = False
    while done is False:
        status, done = downloader.next_chunk()
    fh.seek(0)
    content = fh.read().decode('utf-8')
    return content

def set_public_permission(service, file_id):
    service.permissions().create(
        fileId=file_id,
        body={"type": "anyone", "role": "reader"},
        fields='id',
    ).execute()

def update_file_on_drive(service, file_id, file_content):
    """
    更新Google Drive上的文件内容。
    
    参数:
    - service: Google Drive API服务实例。
    - file_id: 要更新的文件的ID。
    - file_content: 新的文件内容，字符串格式。
    """
    # 将新的文件内容转换为字节流
    fh = io.BytesIO(file_content.encode('utf-8'))
    media = MediaIoBaseUpload(fh, mimetype='application/json', resumable=True)
    
    # 更新文件
    updated_file = service.files().update(
        fileId=file_id,
        media_body=media
    ).execute()
    
    print(f"文件已更新，文件ID: {updated_file['id']}")

# ---- Text file ----
def process_file(password, file):
    verify_password(password)
    
    # 读取文件
    if file.name.endswith('.csv'):
        df = pd.read_csv(file)
        text = df_to_text(df)
    elif file.name.endswith('.xlsx'):
        df = pd.read_excel(file)
        text = df_to_text(df)
    elif file.name.endswith('.docx'):
        text = docx_to_text(file)
    else:
        raise ValueError("Unsupported file type")

    df_string = df.to_string()
    # 宜蘭：移除@XX@符号 to |
    df_string = df_string.replace("@XX@", "|")

    # 根据上传的文件内容生成问题
    questions = generate_questions(df_string)
    summary = generate_summarise(df_string)

    # 返回按钮文本和 DataFrame 字符串
    return questions[0] if len(questions) > 0 else "", \
           questions[1] if len(questions) > 1 else "", \
           questions[2] if len(questions) > 2 else "", \
           summary, \
           df_string

def df_to_text(df):
    # 将 DataFrame 转换为纯文本
    return df.to_string()

def docx_to_text(file):
    # 将 Word 文档转换为纯文本
    doc = Document(file)
    return "\n".join([para.text for para in doc.paragraphs])


# ---- YouTube link ----
def parse_time(time_str):
    """將時間字符串 'HH:MM:SS' 或 'MM:SS' 轉換為 timedelta 物件。"""
    parts = list(map(int, time_str.split(':')))
    if len(parts) == 3:
        hours, minutes, seconds = parts
    elif len(parts) == 2:
        hours = 0  # 沒有小時部分時，將小時設為0
        minutes, seconds = parts
    else:
        raise ValueError("時間格式不正確，應為 'HH:MM:SS' 或 'MM:SS'")
    return timedelta(hours=hours, minutes=minutes, seconds=seconds)

def format_seconds_to_time(seconds):
    """将秒数格式化为 时:分:秒 的形式"""
    hours = int(seconds // 3600)
    minutes = int((seconds % 3600) // 60)
    seconds = int(seconds % 60)
    return f"{hours:02}:{minutes:02}:{seconds:02}"

def extract_youtube_id(url):
    parsed_url = urlparse(url)
    
    if "youtube.com" in parsed_url.netloc:
        # 对于标准链接，视频ID在查询参数'v'中
        query_params = parse_qs(parsed_url.query)
        return query_params.get("v")[0] if "v" in query_params else None
    elif "youtu.be" in parsed_url.netloc:
        # 对于短链接，视频ID是路径的一部分
        return parsed_url.path.lstrip('/')
    else:
        return None

def get_transcript_by_yt_api(video_id):
    transcript_list = YouTubeTranscriptApi.list_transcripts(video_id)
    languages = []
    for t in transcript_list:
        languages.append(t.language_code)
    
    for language in languages:
        try:
            transcript = YouTubeTranscriptApi.get_transcript(video_id, languages=[language])
            print("===transcript===")
            print(transcript)
            print("===transcript===")
            return transcript  # 成功獲取字幕，直接返回結果
        except NoTranscriptFound:
            continue  # 當前語言的字幕沒有找到，繼續嘗試下一個語言
    return None  # 所有嘗試都失敗，返回None

def generate_transcription_by_gemini(video_id):
    """使用 Google Gemini 生成影片逐字稿"""
    print("====generate_transcription_by_gemini====")
    # 準備 YouTube 影片 URL
    video_url = f"https://www.youtube.com/watch?v={video_id}"
    
    # 初始化 Gemini Pro Vision 模型
    model = vertexai.generative_models.GenerativeModel("gemini-2.0-flash-exp")
    
    # 建立影片部分
    video_part = Part.from_uri(
        uri=video_url,
        mime_type="video/*"
    )

    # 設定提示詞
    prompt = "給我包含時間軸的逐字稿，只需要給我有講話的時間軸跟內容，其他時間軸不需要"

    # 生成逐字稿
    original_transcription = ""
    try:
        response = model.generate_content(
            contents=[video_part, prompt],
            generation_config=vertexai.generative_models.GenerationConfig(
                temperature=1.0,
                top_p=0.95,
                max_output_tokens=8192,
                candidate_count=1
            ),
            stream=False
        )

        original_transcription = response.candidates[0].content.parts[0].text
        

        print("===original_transcription===")
        print(original_transcription)
        print("===original_transcription===")


        # 轉換成 JSON 格式
        transcript_json = convert_transcription_to_json(original_transcription)
        
        if transcript_json:
            return transcript_json
        else:
            raise Exception("無法轉換逐字稿格式")
            
    except Exception as e:
        print(f"生成逐字稿時發生錯誤：{str(e)}")
        return None

def convert_transcription_to_json(original_transcription):
    """
    將原始逐字稿轉換成指定的 JSON 格式
    
    Args:
        original_transcription (str): 原始逐字稿文本
        
    Returns:
        list: 包含逐字稿段落的列表，每個段落包含 text, start, end, duration
    """
    
    # 使用 Vertex AI 來處理轉換
    model = vertexai.generative_models.GenerativeModel("gemini-2.0-flash-exp")
    
    prompt = f"""
    請將以下逐字稿轉換成 JSON 格式:
    {original_transcription}
    
    轉換規則:
    1. 每個段落需包含 text, start, end, duration
    2. 時間格式需轉換為秒數(例如 1:02 轉為 62 秒)
    3. duration 為 end - start 的差值
    4. 回傳格式為 JSON array
    
    範例輸出格式:
    [
        {{
            "text": "在一片無人的森林裡",
            "start": 1,
            "end": 2,
            "duration": 1
        }},
        {{
            "text": "你撿到一張羊皮紙",
            "start": 2,
            "end": 4,
            "duration": 2
        }}
    ]
    
    請直接返回 JSON 格式，不要加入任何說明文字。
    """
    
    try:
        response = model.generate_content(prompt)
        json_str = response.text

        print("===json_str===")
        print(json_str)
        print("===json_str===")
        
        # 移除可能的 markdown 標記
        json_str = json_str.replace("```json", "").replace("```", "").strip()
        
        # 解析 JSON
        transcript_json = json.loads(json_str)
        
        # 驗證格式
        for entry in transcript_json:
            if not all(k in entry for k in ["text", "start", "end", "duration"]):
                raise ValueError("JSON 格式錯誤：缺少必要欄位")
                
        return transcript_json
        
    except Exception as e:
        print(f"轉換逐字稿時發生錯誤：{str(e)}")
        return None

def generate_transcription_by_whisper(video_id):
    youtube_url = f'https://www.youtube.com/watch?v={video_id}'
    codec_name = "mp3"
    outtmpl = f"{OUTPUT_PATH}/{video_id}.%(ext)s"
    ydl_opts = {
        'format': 'bestaudio/best',
        'postprocessors': [{
            'key': 'FFmpegExtractAudio',
            'preferredcodec': codec_name,
            'preferredquality': '192'
        }],
        'outtmpl': outtmpl,
    }

    with yt_dlp.YoutubeDL(ydl_opts) as ydl:
        ydl.download([youtube_url])

    audio_path = f"{OUTPUT_PATH}/{video_id}.{codec_name}"
    full_audio = AudioSegment.from_mp3(audio_path)

    max_part_duration = 10 * 60 * 1000  # 10 minutes
    full_duration = len(full_audio)  # in milliseconds
    parts = math.ceil(full_duration / max_part_duration)
    print(f"parts: {parts}")
    transcription = []

    for i in range(parts):
        print(f"== i: {i}==")
        start_time = i * max_part_duration
        end_time = min((i + 1) * max_part_duration, full_duration)
        print(f"time: {start_time/1000} - {end_time/1000}")
        chunk = full_audio[start_time:end_time]
        chunk_path = f"{OUTPUT_PATH}/{video_id}_part_{i}.{codec_name}"
        chunk.export(chunk_path, format=codec_name)

        try:
            with open(chunk_path, "rb") as chunk_file:
                response = OPEN_AI_CLIENT.audio.transcriptions.create(
                    model="whisper-1",
                    file=chunk_file,
                    response_format="verbose_json",
                    timestamp_granularities=["segment"],
                    prompt="Transcribe the following audio file. if content is chinese, please using 'language: zh-TW' ",
                )
                
                # Adjusting the timestamps for the chunk based on its position in the full audio
                adjusted_segments = [{
                    'text': segment.text,  # 使用 .text 屬性而不是 ['text']
                    'start': math.ceil(segment.start + start_time / 1000.0),  # Converting milliseconds to seconds
                    'end': math.ceil(segment.end + start_time / 1000.0),
                    'duration': math.ceil(segment.end - segment.start)
                } for segment in response.segments]

                transcription.extend(adjusted_segments)
        except Exception as e:
            print(f"Error processing chunk {i}: {str(e)}")

        # Remove temporary chunk files after processing
        os.remove(chunk_path)

    return transcription
 
def get_video_duration(video_id):
    yt = YouTube(f'https://www.youtube.com/watch?v={video_id}')
    try:
        video_duration = yt.length
    except:
        video_duration = None

    print(f"video_duration: {video_duration}")
    
    return video_duration

def process_transcript_and_screenshots_on_gcs(video_id):
    print("====process_transcript_and_screenshots_on_gcs====")
    transcript, exists = get_transcript_from_gcs(video_id)
    if not exists:
        print("==== video transcript is not exists ====")
        try:
            transcript = generate_transcription_by_gemini(video_id)
        except Exception as e:
            print(f"generate_transcription_by_gemini Error generating transcription: {str(e)}")
            # transcript = generate_transcription_by_whisper(video_id)

        upload_transcript_to_gcs(video_id, transcript)

    # 處理截圖
    print("====處理截圖====")
    is_new_transcript = False
    has_tried_download_video = False
    for entry in transcript:
        if 'img_file_id' not in entry:
            # 檢查 OUTPUT_PATH 是否存在 video_id.mp4
            video_path = f'{OUTPUT_PATH}/{video_id}.mp4'
            # 沒有影片以及沒有下載過
            if not os.path.exists(video_path) and not has_tried_download_video:
                try:
                    download_youtube_video(video_id)
                except Exception as e:
                    has_tried_download_video = True
                    print(f"下载视频失败: {str(e)}")
                    
            if os.path.exists(video_path):
                try:
                    screenshot_path = screenshot_youtube_video(video_id, entry['start'])
                    screenshot_blob_name = f"{video_id}/{video_id}_{entry['start']}.jpg"
                    img_file_id = GCS_SERVICE.upload_image_and_get_public_url('video_ai_assistant', screenshot_blob_name, screenshot_path)
                    entry['img_file_id'] = img_file_id
                    print(f"截图已上传到GCS: {img_file_id}")
                    is_new_transcript = True
                except Exception as e:
                    print(f"Error processing screenshot: {str(e)}")
            else:
                entry['img_file_id'] = ""
                print(f"截圖空白")
                is_new_transcript = True
    
    if is_new_transcript:
        print("===更新逐字稿文件===")
        upload_transcript_to_gcs(video_id, transcript)

    return transcript

def get_transcript(video_id):
    print("====get_transcript====")
    transcript, exists = get_transcript_from_gcs(video_id)
    if not exists:
        raise gr.Error("逐字稿文件不存在於GCS中。")

    if any('img_file_id' not in entry for entry in transcript):
        raise gr.Error("Some entries in the transcript do not have an associated img_file_id.")

    print("Transcript is verified with all necessary images.")
    return transcript

def get_transcript_from_gcs(video_id):
    print("Checking for transcript in GCS...")
    bucket_name = 'video_ai_assistant'
    transcript_file_name = f'{video_id}_transcript.json'
    transcript_blob_name = f"{video_id}/{transcript_file_name}"
    # Check if the transcript exists in GCS
    is_transcript_exists = GCS_SERVICE.check_file_exists(bucket_name, transcript_blob_name)
    if is_transcript_exists:
        # Download the transcript if it exists
        transcript_text = GCS_SERVICE.download_as_string(bucket_name, transcript_blob_name)
        return json.loads(transcript_text), True
    else:
        print("No transcript found for video ID:", video_id)
        return None, False

def upload_transcript_to_gcs(video_id, transcript):
    print("Uploading updated transcript to GCS...")
    bucket_name = 'video_ai_assistant'
    transcript_file_name = f'{video_id}_transcript.json'
    transcript_blob_name = f"{video_id}/{transcript_file_name}"
    transcript_text = json.dumps(transcript, ensure_ascii=False, indent=2)
    GCS_SERVICE.upload_json_string(bucket_name, transcript_blob_name, transcript_text)
    print("Transcript uploaded successfully.")

def process_youtube_link(password, link, LLM_model=None):
    verify_password(password)
    video_id = extract_youtube_id(link)

    try:
        if IS_ENV_PROD == "True":
            transcript = get_transcript(video_id)
        else:    
            transcript = process_transcript_and_screenshots_on_gcs(video_id)
    except Exception as e:
        error_msg = f" {video_id} 逐字稿錯誤: {str(e)}"
        print("===process_youtube_link error===")
        print(error_msg)
        raise gr.Error(error_msg)

    original_transcript = json.dumps(transcript, ensure_ascii=False, indent=2)
    formatted_transcript = []
    formatted_simple_transcript =[]
    for entry in transcript:
        start_time = format_seconds_to_time(entry['start'])
        end_time = format_seconds_to_time(entry['start'] + entry['duration'])
        embed_url = get_embedded_youtube_link(video_id, entry['start'])
        img_file_id = entry['img_file_id']
        screenshot_path = img_file_id
        line = {
            "start_time": start_time,
            "end_time": end_time,
            "text": entry['text'],
            "embed_url": embed_url,
            "screenshot_path": screenshot_path
        }
        formatted_transcript.append(line)
        # formatted_simple_transcript 只要 start_time, end_time, text
        simple_line = {
            "start_time": start_time,
            "end_time": end_time,
            "text": entry['text']
        }
        formatted_simple_transcript.append(simple_line)

    # 基于逐字稿生成其他所需的输出
    source = "gcs"
    questions_answers = get_questions_answers(video_id, formatted_simple_transcript, source, LLM_model)
    questions_answers_json = json.dumps(questions_answers, ensure_ascii=False, indent=2)
    summary_json = get_video_id_summary(video_id, formatted_simple_transcript, source, LLM_model)
    summary_text = summary_json["summary"]
    summary = summary_json["summary"]
    key_moments_json = get_key_moments(video_id, formatted_simple_transcript, formatted_transcript, source, LLM_model)
    key_moments = key_moments_json["key_moments"]
    key_moments_text = json.dumps(key_moments, ensure_ascii=False, indent=2)
    key_moments_html = get_key_moments_html(key_moments)
    html_content = format_transcript_to_html(formatted_transcript)
    simple_html_content = format_simple_transcript_to_html(formatted_simple_transcript)
    mind_map_json = get_mind_map(video_id, formatted_simple_transcript, source, LLM_model)
    mind_map = mind_map_json["mind_map"]
    mind_map_html = get_mind_map_html(mind_map)
    reading_passage_json = get_reading_passage(video_id, formatted_simple_transcript, source, LLM_model)
    reading_passage_text = reading_passage_json["reading_passage"]
    reading_passage = reading_passage_json["reading_passage"]
    meta_data = get_meta_data(video_id)
    subject = meta_data["subject"]
    grade = meta_data["grade"]

    # 确保返回与 UI 组件预期匹配的输出
    return video_id, \
        questions_answers_json, \
        original_transcript, \
        summary_text, \
        summary, \
        key_moments_text, \
        key_moments_html, \
        mind_map, \
        mind_map_html, \
        html_content, \
        simple_html_content, \
        reading_passage_text, \
        reading_passage, \
        subject, \
        grade
            
def create_formatted_simple_transcript(transcript):
    formatted_simple_transcript = []
    for entry in transcript:
        start_time = format_seconds_to_time(entry['start'])
        end_time = format_seconds_to_time(entry['start'] + entry['duration'])
        line = {
            "start_time": start_time,
            "end_time": end_time,
            "text": entry['text']
        }
        formatted_simple_transcript.append(line)
    return formatted_simple_transcript

def create_formatted_transcript(video_id, transcript):
    formatted_transcript = []
    for entry in transcript:
        start_time = format_seconds_to_time(entry['start'])
        end_time = format_seconds_to_time(entry['start'] + entry['duration'])
        embed_url = get_embedded_youtube_link(video_id, entry['start'])
        img_file_id = entry['img_file_id']
        screenshot_path = img_file_id
        line = {
            "start_time": start_time,
            "end_time": end_time,
            "text": entry['text'],
            "embed_url": embed_url,
            "screenshot_path": screenshot_path
        }
        formatted_transcript.append(line)
    return formatted_transcript

def format_transcript_to_html(formatted_transcript):
    html_content = ""
    for entry in formatted_transcript:
        html_content += f"<h3>{entry['start_time']} - {entry['end_time']}</h3>"
        html_content += f"<p>{entry['text']}</p>"
        html_content += f"<img src='{entry['screenshot_path']}' width='500px' />"
    return html_content

def format_simple_transcript_to_html(formatted_transcript):
    html_content = ""
    for entry in formatted_transcript:
        html_content += f"<h3>{entry['start_time']} - {entry['end_time']}</h3>"
        html_content += f"<p>{entry['text']}</p>"
    return html_content

def get_embedded_youtube_link(video_id, start_time):
    int_start_time = int(start_time)
    embed_url = f"https://www.youtube.com/embed/{video_id}?start={int_start_time}&autoplay=1"
    return embed_url

def download_youtube_video(youtube_id, output_path=OUTPUT_PATH):
    # Construct the full YouTube URL
    youtube_url = f'https://www.youtube.com/watch?v={youtube_id}'

    # Create the output directory if it doesn't exist
    if not os.path.exists(output_path):
        os.makedirs(output_path)
    
    # Download the video
    try:
        yt = YouTube(youtube_url)
        video_stream = yt.streams.filter(progressive=True, file_extension='mp4').order_by('resolution').desc().first()
        video_stream.download(output_path=output_path, filename=youtube_id+".mp4")
        print(f"[Pytube] Video downloaded successfully: {output_path}/{youtube_id}.mp4")
    except Exception as e:
        ydl_opts = {
            'format': "bestvideo[height<=720][ext=mp4]",
            'outtmpl': os.path.join(output_path, f'{youtube_id}.mp4'),  # Output filename template
        }
        with yt_dlp.YoutubeDL(ydl_opts) as ydl:
            ydl.download([youtube_url])

        print(f"[yt_dlp] Video downloaded successfully: {output_path}/{youtube_id}.mp4")

def screenshot_youtube_video(youtube_id, snapshot_sec):    
    video_path = f'{OUTPUT_PATH}/{youtube_id}.mp4'
    file_name = f"{youtube_id}_{snapshot_sec}.jpg"
    with VideoFileClip(video_path) as video:
        screenshot_path = f'{OUTPUT_PATH}/{file_name}'
        video.save_frame(screenshot_path, snapshot_sec)
    
    return screenshot_path


# ---- Web ----
# def process_web_link(link):
#     # 抓取和解析网页内容
#     response = requests.get(link)
#     soup = BeautifulSoup(response.content, 'html.parser')
#     return soup.get_text()


# ---- LLM Generator ----
def split_data(df_string, word_base=100000):
    """Split the JSON string based on a character length base and then chunk the parsed JSON array."""
    if isinstance(df_string, str):
        data_str_cnt = len(df_string)
        data = json.loads(df_string)
    else:
        data_str_cnt = len(str(df_string))
        data = df_string
        
    # Calculate the number of parts based on the length of the string
    n_parts = data_str_cnt // word_base + (1 if data_str_cnt % word_base != 0 else 0)
    print(f"Number of Parts: {n_parts}")
    
    # Calculate the number of elements each part should have
    part_size = len(data) // n_parts if n_parts > 0 else len(data)
    
    segments = []
    for i in range(n_parts):
        start_idx = i * part_size
        end_idx = min((i + 1) * part_size, len(data))
        # Serialize the segment back to a JSON string
        segment = json.dumps(data[start_idx:end_idx]).encode('utf-8').decode('unicode_escape')
        segments.append(segment)
    return segments

def generate_content_by_open_ai(sys_content, user_content, response_format=None, model_name=None):
    print("generate_content_by_open_ai")
    if model_name == "gpt-4-turbo":
        model = "gpt-4-turbo"
    else:
        model = "gpt-4o"
    print(f"LLM model: {model}")

    messages = [
        {"role": "system", "content": sys_content},
        {"role": "user", "content": user_content}
    ]
    request_payload = {
        "model": model,
        "messages": messages,
        "max_tokens": 4000,
    }

    if response_format is not None:
        request_payload["response_format"] = response_format

    response = OPEN_AI_CLIENT.chat.completions.create(**request_payload)
    content = response.choices[0].message.content.strip()
    return content

# def generate_content_by_bedrock(sys_content, user_content):
#     print("LLM using REDROCK")
#     messages = [
#         {"role": "user", "content": user_content +"(如果是 JSON 格式，value 的引號，請用單引號，或是用反斜線＋雙引號，避免 JSON Decoder error )"}
#     ]
#     model_id = "anthropic.claude-3-sonnet-20240229-v1:0"
#     print(f"model_id: {model_id}")
#     # model_id = "anthropic.claude-3-haiku-20240307-v1:0"
#     kwargs = {
#         "modelId": model_id,
#         "contentType": "application/json",
#         "accept": "application/json",
#         "body": json.dumps({
#             "anthropic_version": "bedrock-2023-05-31",
#             "max_tokens": 4000,
#             "system": sys_content,
#             "messages": messages
#         })
#     }
#     response = BEDROCK_CLIENT.invoke_model(**kwargs)
#     response_body = json.loads(response.get('body').read())
#     content = response_body.get('content')[0].get('text')
#     return content

def generate_content_by_gemini(sys_content, user_content, response_format=None, model_name=None):
    print("generate_content_by_gemini")
    print(f"LLM using: {model_name}")

    gemini_model = GenerativeModel(model_name=model_name)
    model_response = gemini_model.generate_content(
        f"{sys_content}, {user_content}"
    )
    content = model_response.candidates[0].content.parts[0].text
    return content


def generate_content_by_LLM(sys_content, user_content, response_format=None, LLM_model=None, model_name=None):
    # 使用 OpenAI 生成基于上传数据的问题

    if LLM_model in ["gemini-1.5-pro","gemini-1.5-flash"]:
        print(f"LLM: {LLM_model}")
        model_name = LLM_model
        content = generate_content_by_gemini(sys_content, user_content, response_format, model_name=model_name)
    # elif LLM_model == "anthropic-claude-3-sonnet":
    #     print(f"LLM: {LLM_model}")
    #     content = generate_content_by_bedrock(sys_content, user_content)
    else:
        print(f"LLM: {LLM_model}")
        print(f"model_name: {model_name}")
        content = generate_content_by_open_ai(sys_content, user_content, response_format, model_name=model_name)

    print("=====content=====")
    print(content)
    print("=====content=====")

    return content

def get_reading_passage(video_id, df_string, source, LLM_model=None):
    if source == "gcs":
        print("===get_reading_passage on gcs===")
        bucket_name = 'video_ai_assistant'
        file_name = f'{video_id}_reading_passage_latex.json'
        blob_name = f"{video_id}/{file_name}"
        # 检查 reading_passage 是否存在
        is_file_exists = GCS_SERVICE.check_file_exists(bucket_name, blob_name)
        if not is_file_exists:
            reading_passage = generate_reading_passage(df_string, LLM_model)
            reading_passage_json = {"reading_passage": str(reading_passage)}
            reading_passage_text = json.dumps(reading_passage_json, ensure_ascii=False, indent=2)
            GCS_SERVICE.upload_json_string(bucket_name, blob_name, reading_passage_text)
            print("reading_passage已上传到GCS")
        else:
            # reading_passage已存在，下载内容
            print("reading_passage已存在于GCS中")
            reading_passage_text = GCS_SERVICE.download_as_string(bucket_name, blob_name)
            reading_passage_json = json.loads(reading_passage_text)

    elif source == "drive":
        print("===get_reading_passage on drive===")
        service = init_drive_service()
        parent_folder_id = '1GgI4YVs0KckwStVQkLa1NZ8IpaEMurkL'
        folder_id = create_folder_if_not_exists(service, video_id, parent_folder_id)
        file_name = f'{video_id}_reading_passage.json'

        # 检查 reading_passage 是否存在
        exists, file_id = check_file_exists(service, folder_id, file_name)
        if not exists:
            reading_passage = generate_reading_passage(df_string)
            reading_passage_json = {"reading_passage": str(reading_passage)}
            reading_passage_text = json.dumps(reading_passage_json, ensure_ascii=False, indent=2)
            upload_content_directly(service, file_name, folder_id, reading_passage_text)
            print("reading_passage已上傳到Google Drive")
        else:
            # reading_passage已存在，下载内容
            print("reading_passage已存在于Google Drive中")
            reading_passage_text = download_file_as_string(service, file_id)

    return reading_passage_json

def generate_reading_passage(df_string, LLM_model=None):
    print("===generate_reading_passage 0===")
    print(df_string)

    segments = split_data(df_string, word_base=100000)
    all_content = []
    model_name = "gpt-4-turbo"
    # model_name = "gpt-4o"

    for segment in segments:
        sys_content = "你是一個擅長資料分析跟影片教學的老師，user 為學生，請精讀資料文本，自行判斷資料的種類，使用 zh-TW"        
        user_content = f"""
            # 文本 {segment}
            
            # rules:
            - 根據文本，抓取重點
            - 去除人類講課時口語的問答句，重新拆解成文章，建立適合閱讀語句通順的 Reading Passage
            - 只需要專注提供 Reading Passage，字數在 500 字以內
            - 敘述中，請把數學或是專業術語，用 Latex 包覆（$...$）
            - 加減乘除、根號、次方等等的運算式口語也換成 LATEX 數學符號
            
            # restrictions:
            - 請一定要使用繁體中文 zh-TW，這很重要
            - 產生的結果不要前後文解釋，也不要敘述這篇文章怎麼產生的
            - 請直接給出文章，不用介紹怎麼處理的或是文章字數等等
            - 字數在 500 字以內  
        """

        print("======user_content 0 ===")
        print(user_content)

        content = generate_content_by_LLM(sys_content, user_content, response_format=None, LLM_model=LLM_model, model_name=model_name)
        all_content.append(content + "\n")

    # 將所有生成的閱讀理解段落合併成一個完整的文章
    final_content = "\n".join(all_content)
    return final_content

def text_to_speech(video_id, text):
    tts = gTTS(text, lang='en')
    filename = f'{video_id}_reading_passage.mp3'
    tts.save(filename)
    return filename

def get_mind_map(video_id, df_string, source, LLM_model=None):
    if source == "gcs":
        print("===get_mind_map on gcs===")
        gcs_client = GCS_CLIENT
        bucket_name = 'video_ai_assistant'
        file_name = f'{video_id}_mind_map.json'
        blob_name = f"{video_id}/{file_name}"
        # 检查檔案是否存在
        is_file_exists = GCS_SERVICE.check_file_exists(bucket_name, blob_name)
        if not is_file_exists:
            mind_map = generate_mind_map(df_string, LLM_model)
            mind_map_json = {"mind_map": str(mind_map)}
            mind_map_text = json.dumps(mind_map_json, ensure_ascii=False, indent=2)
            GCS_SERVICE.upload_json_string(bucket_name, blob_name, mind_map_text)
            print("mind_map已上傳到GCS")
        else:
            # mindmap已存在，下载内容
            print("mind_map已存在于GCS中")
            mind_map_text = GCS_SERVICE.download_as_string(bucket_name, blob_name)
            mind_map_json = json.loads(mind_map_text)

    elif source == "drive":
        print("===get_mind_map on drive===")
        service = init_drive_service()
        parent_folder_id = '1GgI4YVs0KckwStVQkLa1NZ8IpaEMurkL'
        folder_id = create_folder_if_not_exists(service, video_id, parent_folder_id)
        file_name = f'{video_id}_mind_map.json'

        # 检查檔案是否存在
        exists, file_id = check_file_exists(service, folder_id, file_name)
        if not exists:
            mind_map = generate_mind_map(df_string, LLM_model)
            mind_map_json = {"mind_map": str(mind_map)}
            mind_map_text = json.dumps(mind_map_json, ensure_ascii=False, indent=2)
            upload_content_directly(service, file_name, folder_id, mind_map_text)
            print("mind_map已上傳到Google Drive")
        else:
            # mindmap已存在，下载内容
            print("mind_map已存在于Google Drive中")
            mind_map_text = download_file_as_string(service, file_id)
            mind_map_json = json.loads(mind_map_text)

    return mind_map_json

def generate_mind_map(df_string, LLM_model=None):
    print("===generate_mind_map===")
    segments = split_data(df_string, word_base=100000)
    all_content = []

    for segment in segments:
        sys_content = "你是一個擅長資料分析跟影片教學的老師，user 為學生，請精讀資料文本，自行判斷資料的種類，使用 zh-TW"
        user_content = f"""
            請根據 {segment} 文本建立 markdown 心智圖
            注意：不需要前後文敘述，直接給出 markdown 文本即可
            這對我很重要
        """
        content = generate_content_by_LLM(sys_content, user_content, response_format=None, LLM_model=LLM_model, model_name=None)
        all_content.append(content + "\n")

    # 將所有生成的閱讀理解段落合併成一個完整的文章
    final_content = "\n".join(all_content)
    return final_content

def get_mind_map_html(mind_map):
    mind_map_markdown = mind_map.replace("```markdown", "").replace("```", "")

    mind_map_html = f"""
    <div class="markmap">
        <script type="text/template">
            {mind_map_markdown}
        </script>
    </div>
    """
    return mind_map_html

def get_video_id_summary(video_id, df_string, source, LLM_model=None):
    if source == "gcs":
        print("===get_video_id_summary on gcs===")
        bucket_name = 'video_ai_assistant'
        file_name = f'{video_id}_summary_markdown.json'
        summary_file_blob_name = f"{video_id}/{file_name}"
        # 检查 summary_file 是否存在
        is_summary_file_exists = GCS_SERVICE.check_file_exists(bucket_name, summary_file_blob_name)
        if not is_summary_file_exists:
            meta_data = get_meta_data(video_id)
            summary = generate_summarise(df_string, meta_data, LLM_model)
            summary_json = {"summary": str(summary)}
            summary_text = json.dumps(summary_json, ensure_ascii=False, indent=2)
            GCS_SERVICE.upload_json_string(bucket_name, summary_file_blob_name, summary_text)
            print("summary已上传到GCS")
        else:
            # summary已存在，下载内容
            print("summary已存在于GCS中")
            summary_text = GCS_SERVICE.download_as_string(bucket_name, summary_file_blob_name)
            summary_json = json.loads(summary_text)

    elif source == "drive":
        print("===get_video_id_summary===")
        service = init_drive_service()
        parent_folder_id = '1GgI4YVs0KckwStVQkLa1NZ8IpaEMurkL'
        folder_id = create_folder_if_not_exists(service, video_id, parent_folder_id)
        file_name = f'{video_id}_summary.json'

        # 检查逐字稿是否存在
        exists, file_id = check_file_exists(service, folder_id, file_name)
        if not exists:
            meta_data = get_meta_data(video_id)
            summary = generate_summarise(df_string, meta_data, LLM_model)
            summary_json = {"summary": str(summary)}
            summary_text = json.dumps(summary_json, ensure_ascii=False, indent=2)
            
            try:
                upload_content_directly(service, file_name, folder_id, summary_text)
                print("summary已上傳到Google Drive")

            except Exception as e:
                error_msg = f" {video_id} 摘要錯誤: {str(e)}"
                print("===get_video_id_summary error===")
                print(error_msg)
                print("===get_video_id_summary error===")
        else:
            # 逐字稿已存在，下载逐字稿内容
            print("summary已存在Google Drive中")
            summary_text = download_file_as_string(service, file_id)
            summary_json = json.loads(summary_text)
        
    return summary_json
    
def generate_summarise(df_string, metadata=None, LLM_model=None):
    print("===generate_summarise===")
    # 使用 OpenAI 生成基于上传数据的问题
    if metadata:
        title = metadata.get("title", "")
        subject = metadata.get("subject", "")
        grade = metadata.get("grade", "")
    else:
        title = ""
        subject = ""
        grade = ""

    segments = split_data(df_string, word_base=100000)
    all_content = []

    for segment in segments:
        sys_content = "你是一個擅長資料分析跟影片教學的老師，user 為學生，請精讀資料文本，自行判斷賛料的種類，使用 zh-TW"
        user_content = f"""
            課程名稱：{title}
            科目：{subject}
            年級：{grade}

            請根據內文： {segment}
            
            格式為 Markdown
            如果有課程名稱，請圍繞「課程名稱」為學習重點，進行重點整理，不要整理跟情境故事相關的問題
            整體摘要在一百字以內
            重點概念列出 bullet points，至少三個，最多五個
            以及可能的結論與結尾延伸小問題提供學生作反思
            敘述中，請把數學或是專業術語，用 Latex 包覆（$...$）
            加減乘除、根號、次方等等的運算式口語也換成 LATEX 數學符號

            整體格式為：
            ## 🌟 主題：{{title}} (如果沒有 title 就省略)
            ## 📚 整體摘要
                - (一個 bullet point....)

            ## 🔖 重點概念
                - xxx
                - xxx
                - xxx
            
            ## 💡 為什麼我們要學這個？
                - (一個 bullet point....)
            
            ## ❓ 延伸小問題
                - (一個 bullet point....請圍繞「課程名稱」為學習重點，進行重點整理，不要整理跟情境故事相關的問題)
        """
        content = generate_content_by_LLM(sys_content, user_content, response_format=None, LLM_model=LLM_model, model_name=None)
        all_content.append(content + "\n")

    if len(all_content) > 1:
        all_content_cnt = len(all_content)
        all_content_str = json.dumps(all_content)
        sys_content = "你是一個擅長資料分析跟影片教學的老師，user 為學生，請精讀賛料文本，自行判斷賛料的種類，使用 zh-TW"
        user_content = f"""
            課程名稱：{title}
            科目：{subject}
            年級：{grade}

            請根據內文： {all_content_str}
            共有 {all_content_cnt} 段，請縱整成一篇摘要
            
            格式為 Markdown
            如果有課程名稱，請圍繞「課程名稱」為學習重點，進行重點整理，不要整理跟情境故事相關的問題
            整體摘要在 {all_content_cnt} 百字以內
            重點概念列出 bullet points，至少三個，最多十個
            以及可能的結論與結尾延伸小問題提供學生作反思
            敘述中，請把數學或是專業術語，用 Latex 包覆（$...$）
            加減乘除、根號、次方等等的運算式口語也換成 LATEX 數學符號

            整體格式為：
            ## 🌟 主題：{{title}} (如果沒有 title 就省略)
            ## 📚 整體摘要
                - ( {all_content_cnt} 個 bullet point....)

            ## 🔖 重點概念
                - xxx
                - xxx
                - xxx
            
            ## 💡 為什麼我們要學這個？
                - ( {all_content_cnt}  個 bullet point....)
            
            ## ❓ 延伸小問題
                - ( {all_content_cnt}  個 bullet point....請圍繞「課程名稱」為學習重點，進行重點整理，不要整理跟情境故事相關的問題)
        """
        final_content = generate_content_by_LLM(sys_content, user_content, response_format=None, LLM_model=LLM_model, model_name=None)
    else:
        final_content = all_content[0]

    return final_content

def get_questions(video_id, df_string, source="gcs", LLM_model=None):
    if source == "gcs":
        # 去 gcs 確認是有有 video_id_questions.json
        print("===get_questions on gcs===")
        gcs_client = GCS_CLIENT
        bucket_name = 'video_ai_assistant'
        file_name = f'{video_id}_questions.json'
        blob_name = f"{video_id}/{file_name}"
        # 检查檔案是否存在
        is_questions_exists = GCS_SERVICE.check_file_exists(bucket_name, blob_name)
        if not is_questions_exists:
            questions = generate_questions(df_string, LLM_model)
            questions_text = json.dumps(questions, ensure_ascii=False, indent=2)
            GCS_SERVICE.upload_json_string(bucket_name, blob_name, questions_text)
            print("questions已上傳到GCS")
        else:
            # 逐字稿已存在，下载逐字稿内容
            print("questions已存在于GCS中")
            questions_text = GCS_SERVICE.download_as_string(bucket_name, blob_name)
            questions = json.loads(questions_text)
    
    elif source == "drive":
        # 去 g drive 確認是有有 video_id_questions.json
        print("===get_questions===")
        service = init_drive_service()
        parent_folder_id = '1GgI4YVs0KckwStVQkLa1NZ8IpaEMurkL'
        folder_id = create_folder_if_not_exists(service, video_id, parent_folder_id)
        file_name = f'{video_id}_questions.json'

        # 检查檔案是否存在
        exists, file_id = check_file_exists(service, folder_id, file_name)
        if not exists:
            questions = generate_questions(df_string, LLM_model)
            questions_text = json.dumps(questions, ensure_ascii=False, indent=2)
            upload_content_directly(service, file_name, folder_id, questions_text)
            print("questions已上傳到Google Drive")
        else:
            # 逐字稿已存在，下载逐字稿内容
            print("questions已存在于Google Drive中")
            questions_text = download_file_as_string(service, file_id)
            questions = json.loads(questions_text)

    q1 = questions[0] if len(questions) > 0 else ""
    q2 = questions[1] if len(questions) > 1 else ""
    q3 = questions[2] if len(questions) > 2 else ""
    print("=====get_questions=====")
    print(f"q1: {q1}")
    print(f"q2: {q2}")
    print(f"q3: {q3}")
    print("=====get_questions=====")
    return q1, q2, q3

def generate_questions(df_string, LLM_model=None):
    print("===generate_questions===")
    # 使用 OpenAI 生成基于上传数据的问题
    if isinstance(df_string, str):
        df_string_json = json.loads(df_string)
    else:
        df_string_json = df_string
    content_text = ""

    for entry in df_string_json:
        content_text += entry["text"] + "，"

    sys_content = "你是一個擅長資料分析跟影片教學的老師，user 為學生，請精讀資料文本，自行判斷資料的種類，並用既有資料為本質猜測用戶可能會問的問題，使用 zh-TW"
    user_content = f"""
        請根據 {content_text} 生成三個問題，並用 JSON 格式返回
        一定要使用 zh-TW，這非常重要！
        
        EXAMPLE:
        {{
            questions:
            [q1的敘述text, q2的敘述text, q3的敘述text]
        }}
    """
    response_format = { "type": "json_object" }
    questions = generate_content_by_LLM(sys_content, user_content, response_format, LLM_model, model_name=None)
    questions_list = json.loads(questions)["questions"]
    print("=====json_response=====")
    print(questions_list)
    print("=====json_response=====")

    return questions_list

def get_questions_answers(video_id, df_string, source="gcs", LLM_model=None):
    if source == "gcs":
        try:
            print("===get_questions_answers on gcs===")
            bucket_name = 'video_ai_assistant'
            file_name = f'{video_id}_questions_answers.json'
            blob_name = f"{video_id}/{file_name}"
            # 检查檔案是否存在
            is_questions_answers_exists = GCS_SERVICE.check_file_exists(bucket_name, blob_name)
            if not is_questions_answers_exists:
                questions_answers = generate_questions_answers(df_string, LLM_model)
                questions_answers_text = json.dumps(questions_answers, ensure_ascii=False, indent=2)
                GCS_SERVICE.upload_json_string(bucket_name, blob_name, questions_answers_text)
                print("questions_answers已上傳到GCS")
            else:
                # questions_answers已存在，下载内容
                print("questions_answers已存在于GCS中")
                questions_answers_text = GCS_SERVICE.download_as_string(bucket_name, blob_name)
                questions_answers = json.loads(questions_answers_text)
        except Exception as e:
            print(f"Error getting questions_answers: {str(e)}")
            questions_list = get_questions(video_id, df_string, source, LLM_model)
            questions_answers = [{"question": q, "answer": ""} for q in questions_list]

    return questions_answers

def generate_questions_answers(df_string, LLM_model=None):
    print("===generate_questions_answers===")
    segments = split_data(df_string, word_base=100000)
    all_content = []

    for segment in segments:
        sys_content = "你是一個擅長資料分析跟影片教學的老師，user 為學生，請精讀資料文本，自行判斷資料的種類，使用 zh-TW"
        user_content = f"""
            請根據 {segment} 生成三個問題跟答案，主要與學科有關，不要問跟情節故事相關的問題
            答案要在最後標示出處【參考：00:01:05】，請根據時間軸 start_time 來標示
            請確保問題跟答案都是繁體中文 zh-TW
            答案不用是標準答案，而是帶有啟發性的蘇格拉底式問答，讓學生思考本來的問題，以及該去參考的時間點
            並用 JSON 格式返回 list ，請一定要給三個問題跟答案，且要裝在一個 list 裡面
            k-v pair 的 key 是 question, value 是 answer

            EXAMPLE:
            {{
            "questions_answers":
            [
                {{question: q1的敘述text, answer: q1的答案text【參考：00:01:05】}}, 
                {{question: q2的敘述text, answer: q2的答案text【參考：00:32:05】}},
                {{question: q3的敘述text, answer: q3的答案text【參考：01:03:35】}}
            ]
            }}
        """
        response_format = { "type": "json_object" }
        content = generate_content_by_LLM(sys_content, user_content, response_format, LLM_model, model_name=None)
        content_json = json.loads(content)["questions_answers"]
        all_content += content_json

        print("=====all_content=====")
        print(all_content)
        print("=====all_content=====")

    return all_content

def change_questions(password, df_string):
    verify_password(password)

    questions = generate_questions(df_string)
    q1 = questions[0] if len(questions) > 0 else ""
    q2 = questions[1] if len(questions) > 1 else ""
    q3 = questions[2] if len(questions) > 2 else ""
    print("=====get_questions=====")
    print(f"q1: {q1}")
    print(f"q2: {q2}")
    print(f"q3: {q3}")
    print("=====get_questions=====")
    return q1, q2, q3

def get_key_moments(video_id, formatted_simple_transcript, formatted_transcript, source, LLM_model=None):
    if source == "gcs":
        print("===get_key_moments on gcs===")
        bucket_name = 'video_ai_assistant'
        file_name = f'{video_id}_key_moments.json'
        blob_name = f"{video_id}/{file_name}"
        # 检查檔案是否存在
        is_key_moments_exists = GCS_SERVICE.check_file_exists(bucket_name, blob_name)
        if not is_key_moments_exists:
            key_moments = generate_key_moments(formatted_simple_transcript, formatted_transcript, LLM_model)
            key_moments_json = {"key_moments": key_moments}
            key_moments_text = json.dumps(key_moments_json, ensure_ascii=False, indent=2)
            GCS_SERVICE.upload_json_string(bucket_name, blob_name, key_moments_text)
            print("key_moments已上傳到GCS")
        else:
            # key_moments已存在，下载内容
            print("key_moments已存在于GCS中")
            key_moments_text = GCS_SERVICE.download_as_string(bucket_name, blob_name)
            key_moments_json = json.loads(key_moments_text)
             # 檢查 key_moments 是否有 keywords
            print("===檢查 key_moments 是否有 keywords===")
            has_keywords_added = False
            for key_moment in key_moments_json["key_moments"]:
                if "keywords" not in key_moment:
                    transcript = key_moment["transcript"]
                    key_moment["keywords"] = generate_key_moments_keywords(transcript, LLM_model)
                    print("===keywords===")
                    print(key_moment["keywords"])
                    print("===keywords===")
                    has_keywords_added = True
            if has_keywords_added:
                key_moments_text = json.dumps(key_moments_json, ensure_ascii=False, indent=2)
                GCS_SERVICE.upload_json_string(bucket_name, blob_name, key_moments_text)
                key_moments_text = GCS_SERVICE.download_as_string(bucket_name, blob_name)
                key_moments_json = json.loads(key_moments_text)
            # 檢查 key_moments 是否有 suggested_images
            print("===檢查 key_moments 是否有 suggested_images===")
            has_suggested_images_added = False
            for key_moment in key_moments_json["key_moments"]:
                if "suggested_images" not in key_moment:
                    key_moment["suggested_images"] = generate_key_moments_suggested_images(key_moment)
                    has_suggested_images_added = True
            if has_suggested_images_added:
                key_moments_text = json.dumps(key_moments_json, ensure_ascii=False, indent=2)
                GCS_SERVICE.upload_json_string(bucket_name, blob_name, key_moments_text)
                key_moments_text = GCS_SERVICE.download_as_string(bucket_name, blob_name)
                key_moments_json = json.loads(key_moments_text)

    elif source == "drive":
        print("===get_key_moments on drive===")
        service = init_drive_service()
        parent_folder_id = '1GgI4YVs0KckwStVQkLa1NZ8IpaEMurkL'
        folder_id = create_folder_if_not_exists(service, video_id, parent_folder_id)
        file_name = f'{video_id}_key_moments.json'

        # 检查檔案是否存在
        exists, file_id = check_file_exists(service, folder_id, file_name)
        if not exists:
            key_moments = generate_key_moments(formatted_simple_transcript, formatted_transcript, LLM_model)
            key_moments_json = {"key_moments": key_moments}
            key_moments_text = json.dumps(key_moments_json, ensure_ascii=False, indent=2)
            upload_content_directly(service, file_name, folder_id, key_moments_text)
            print("key_moments已上傳到Google Drive")
        else:
            # key_moments已存在，下载内容
            print("key_moments已存在于Google Drive中")
            key_moments_text = download_file_as_string(service, file_id)
            key_moments_json = json.loads(key_moments_text)
    
    return key_moments_json

def generate_key_moments(formatted_simple_transcript, formatted_transcript, LLM_model=None):
    print("===generate_key_moments===")
    segments = split_data(formatted_simple_transcript, word_base=100000)
    all_content = []

    for segment in segments:
        sys_content = "你是一個擅長資料分析跟影片教學的老師，user 為學生，請精讀資料文本，自行判斷資料的種類，使用 zh-TW"
        user_content = f"""
            # 文本：{segment} 

            # Rule
            1. 請根據文本，提取出 5~8 段重點摘要，並給出對應的時間軸，每一段重點的時間軸範圍大於1分鐘，但小於 1/3 總逐字稿長度
            2. 內容當中，如果有列舉方法、模式或是工具，就用 bulletpoint 或是 編號方式 列出，並在列舉部分的頭尾用[]匡列（example: FAANG 是以下五間公司： [1. A公司 2.B公司 3.C公司 4.D公司 5.E公司 ]，...）
            3. 注意不要遺漏任何一段時間軸的內容 從零秒開始，以這種方式分析整個文本，從零秒開始分析，直到結束。這很重要
            4. 結尾的時間如果有總結性的話，也要擷取
            5. 如果頭尾的情節不是重點，特別是打招呼或是介紹自己是誰、或是finally say goodbye 就是不重要的情節，就不用擷取
            6. 關鍵字從transcript extract to keyword，保留專家名字、專業術語、年份、數字、期刊名稱、地名、數學公式
            7. 最後再檢查一遍，text, keywords please use or transfer to zh-TW, it's very important

            # restrictions
            1. 請一定要用 zh-TW，這非常重要！
            2. 如果是疑似主播、主持人的圖片場景，且沒有任何有用的資訊，請不要選取
            3. 如果頭尾的情節不是重點，特別是打招呼或是介紹自己是誰、或是finally say goodbye 就是不重要的情節，就不用擷取
            4. 時間軸請取到秒數，不要只取到分鐘數，這很重要

            Example: retrun JSON 
            {{key_moments:[{{
                "start": "00:00", 
                "end": "01:35", 
                "text": "逐字稿的重點摘要",
                "keywords": ["關鍵字", "關鍵字"]
                }}]
            }}
        """
        response_format = { "type": "json_object" }
        content = generate_content_by_LLM(sys_content, user_content, response_format, LLM_model, model_name=None)
        key_moments = json.loads(content)["key_moments"]

        # "transcript": get text from formatted_simple_transcript
        for moment in key_moments:
            start_time = parse_time(moment['start'])
            end_time = parse_time(moment['end'])
            # 使用轉換後的 timedelta 物件進行時間
            moment['transcript'] = "，".join([entry['text'] for entry in formatted_simple_transcript
                                            if start_time <= parse_time(entry['start_time']) <= end_time])

        print("=====key_moments=====")
        print(key_moments)
        print("=====key_moments=====")
        image_links = {entry['start_time']: entry['screenshot_path'] for entry in formatted_transcript}

        for moment in key_moments:
            start_time = parse_time(moment['start'])
            end_time = parse_time(moment['end'])
            # 使用轉換後的 timedelta 物件進行時間比較
            moment_images = [image_links[time] for time in image_links 
                            if start_time <= parse_time(time) <= end_time]
            moment['images'] = moment_images

            # 檢查是否有 suggested_images
            if "suggested_images" not in moment:
                moment["suggested_images"] = generate_key_moments_suggested_images(moment, LLM_model)
                print("===moment_suggested_images===")
                print(moment["suggested_images"])
                print("===moment_suggested_images===")

        all_content += key_moments

    return all_content

def generate_key_moments_keywords(transcript, LLM_model=None):
    print("===generate_key_moments_keywords===")
    segments = split_data(transcript, word_base=100000)
    all_content = []

    for segment in segments:
        sys_content = "你是一個擅長資料分析跟影片教學的老師，user 為學生，請精讀資料文本，自行判斷資料的種類，使用 zh-TW"
        user_content = f"""
            transcript  extract to keyword
            保留專家名字、專業術語、年份、數字、期刊名稱、地名、數學公式、數學表示式、物理化學符號，
            不用給上下文，直接給出關鍵字，使用 zh-TW，用逗號分隔， example: 關鍵字1, 關鍵字2
            transcript：{segment}
        """
        content = generate_content_by_LLM(sys_content, user_content, response_format=None, LLM_model=LLM_model, model_name=None)
        keywords = content.strip().split(",")
        all_content += keywords
    
    return all_content

def generate_key_moments_suggested_images(key_moment, LLM_model=None):
    # Prepare the text and keywords
    text = key_moment["text"]
    keywords = ', '.join(key_moment["keywords"])
    images = key_moment["images"]

    images_list_prompt = ""
    for i, image_url in enumerate(images):
        images_list_prompt += f"\n圖片 {i+1}: {image_url}"

    # Prepare the user prompt with text and keywords
    sys_content = "你是一個擅長資料分析跟影片教學的老師，user 為學生，請精讀資料文本，自行判斷資料的種類，使用 zh-TW"
    user_content = f"""
    - 文本: {text}
    - 關鍵字: {keywords}
    
    # Rule:
    1. 保留有圖表或是數據的圖片
    2. 根據文本和關鍵字，選擇出最合適的圖片。
    3. 總是保留最後一張，除非他是一張空白圖片，或是一張沒有任何內容的圖片
    
    # Restrictions: 
    1. 如果是疑似主播、主持人的圖片場景，且沒有任何有用的資訊，請不要選取，這很重要
    2. 不要有相似或是概念重複的圖片
    3. 移除整張圖片是黑色、藍色或是白色的圖片
    4. 移除沒有任何內容的圖片
    5. 不需要理會字幕的差益，只需要看圖片的內容

    請根據這些信息，圖片列表如下:
    {images_list_prompt}

    回傳 JSON LIST 就好，不用回傳任何敘述脈絡，也不要 ```json 包覆
    EXAMPLE:
    {{
        "suggested_images": ["圖片1的 image_url", "圖片2 的 image_url", "圖片3的 image_url"]
    }}
    """

    response_format = { "type": "json_object" }
    response = generate_content_by_LLM(sys_content, user_content, response_format, LLM_model, model_name=None)
    print("===generate_key_moments_suggested_images===")
    print(response)
    print("===generate_key_moments_suggested_images===")
    suggested_images = json.loads(response)["suggested_images"]

    return suggested_images

def get_key_moments_html(key_moments):
    css = """
    <style>
    #gallery-main {
        display: flex;
        align-items: center;
        margin-bottom: 20px;
    }

    #gallery {
        position: relative;
        width: 50%;
        flex: 1;
    }

    #text-content {
        flex: 2;
        margin-left: 20px;
    }


    #gallery #gallery-container{
        position: relative;
        width: 100%;
        height: 0px;
        padding-bottom: 56.7%; /* 16/9 ratio */
        background-color: blue;
    }
    #gallery #gallery-container #gallery-content{
        position: absolute;
        top: 0px;
        right: 0px;
        bottom: 0px;
        left: 0px;
        height: 100%;
        display: flex;
        scroll-snap-type: x mandatory;
        overflow-x: scroll;
        scroll-behavior: smooth;
    }
    #gallery #gallery-container #gallery-content .gallery__item{
        width: 100%;
        height: 100%;
        flex-shrink: 0;
        scroll-snap-align: start;
        scroll-snap-stop: always;
        position: relative;
    }
    #gallery #gallery-container #gallery-content .gallery__item img{
        display: block;
        width: 100%;
        height: 100%;
        object-fit: contain;
        background-color: white;
    }

    .click-zone{
        position: absolute;
        width: 20%;
        height: 100%;
        z-index: 3;
    }
    .click-zone.click-zone-prev{
        left: 0px;
    }
    .click-zone.click-zone-next{
        right: 0px;
    }
    #gallery:not(:hover) .arrow{
        opacity: 0.8;
    }
    .arrow{
        text-align: center;
        z-index: 3;
        position: absolute;
        display: block;
        width: 25px;
        height: 25px;
        line-height: 25px;
        background-color: black;
        border-radius: 50%;
        text-decoration: none;
        color: white !important;
        opacity: 0.8; 
        transition: opacity 200ms ease;
    }
    .arrow:hover{
        opacity: 1;
    }
    .arrow span{
        position: relative;
        top: 2px;
    }
    .arrow.arrow-prev{
        top: 50%;
        left: 5px;
    }
    .arrow.arrow-next{
        top: 50%;
        right: 5px;
    }
    .arrow.arrow-disabled{
        opacity:0.8;
    }

    #text-content {
        padding: 0px 36px;
    }
    #text-content p {
        margin-top: 10px;
    }

    body{
        font-family: sans-serif;
        margin: 0px;
        padding: 0px;
    }
    main{
        padding: 0px;
        margin: 0px;
        max-width: 900px;
        margin: auto;
    }
    .hidden{
        border: 0;
        clip: rect(0 0 0 0);
        height: 1px;
        margin: -1px;
        overflow: hidden;
        padding: 0;
        position: absolute;
        width: 1px;
    }
    .keyword-label {
        display: inline-block;
        padding: 5px 10px;
        margin: 2px;
        border: 2px solid black;
        border-radius: 5px;
        font-size: 0.9em;
    }
    details {
        border-radius: 5px;
        padding: 10px;
        border: 2px solid black;
    }

    summary {
        font-weight: bold;
        cursor: pointer;
        outline: none;
    }

    summary::-webkit-details-marker {
        display: none;
    }

    @media (max-width: 768px) {
        #gallery-main {
            flex-direction: column; /* 在小屏幕上堆叠元素 */
        }

        #gallery {
            width: 100%; /* 让画廊占满整个容器宽度 */
        }

        #text-content {
            margin-left: 0; /* 移除左边距，让文本内容占满宽度 */
            margin-top: 20px; /* 为文本内容添加顶部间距 */
        }

        #gallery #gallery-container {
            height: 350px; /* 或者你可以设置一个固定的高度，而不是用 padding-bottom */
            padding-bottom: 0; /* 移除底部填充 */
        }
    }
    </style>
    """

    key_moments_html = css

    for i, moment in enumerate(key_moments):
        # if "suggested_images" in moment:
        #     images = moment['suggested_images']
        # else:
        #     images = moment['images']
        # image_elements = ""
        
        # for j, image in enumerate(images):
        #     current_id = f"img_{i}_{j}"
        #     prev_id = f"img_{i}_{j-1}" if j-1 >= 0 else f"img_{i}_{len(images)-1}"
        #     next_id = f"img_{i}_{j+1}" if j+1 < len(images) else f"img_{i}_0"

        #     image_elements += f"""
        #         <div id="{current_id}" class="gallery__item">
        #             <a href="#{prev_id}" class="click-zone click-zone-prev">
        #                 <div class="arrow arrow-disabled arrow-prev"> ◀︎ </div>
        #             </a> 
        #             <a href="#{next_id}" class="click-zone click-zone-next">
        #                 <div class="arrow arrow-next"> ▶︎ </div>
        #             </a>
        #             <img src="{image}">
        #         </div>
        #     """

        # gallery_content = f"""
        #     <div id="gallery-content">
        #         {image_elements}
        #     </div>
        # """

        keywords_html = ' '.join([f'<span class="keyword-label">{keyword}</span>' for keyword in moment['keywords']])


        key_moments_html += f"""
        <div class="gallery-container" id="gallery-main">
            <div id="text-content">
                <h3>{moment['start']} - {moment['end']}</h3>
                <p><strong>摘要: {moment['text']} </strong></p>
                <details>
                    <summary>逐字稿</summary>
                    <p><strong>內容: </strong> {moment['transcript']} </p>
                </details>
                <p><strong>關鍵字:</strong> {keywords_html}</p>
            </div>
        </div>
        """

    return key_moments_html

# ---- LLM CRUD ----
def get_LLM_content(video_id, kind):
    print(f"===get_{kind}===")
    gcs_client = GCS_CLIENT
    bucket_name = 'video_ai_assistant'
    file_name = f'{video_id}_{kind}.json'
    blob_name = f"{video_id}/{file_name}"
    # 检查 file 是否存在
    is_file_exists = GCS_SERVICE.check_file_exists(bucket_name, blob_name)
    if is_file_exists:
        content = GCS_SERVICE.download_as_string(bucket_name, blob_name)
        content_json = json.loads(content)
        if kind == "reading_passage_latex":
            content_text = content_json["reading_passage"]
        elif kind == "summary_markdown":
            content_text = content_json["summary"]
        elif kind == "key_moments":
            content_text = content_json["key_moments"]
            content_text = json.dumps(content_text, ensure_ascii=False, indent=2)
        else:
            content_text = json.dumps(content_json, ensure_ascii=False, indent=2)
    else:
        content_text = ""
    return content_text

def enable_edit_mode():
    return gr.update(interactive=True)

def delete_LLM_content(video_id, kind):
    print(f"===delete_{kind}===")
    gcs_client = GCS_CLIENT
    bucket_name = 'video_ai_assistant'
    file_name = f'{video_id}_{kind}.json'
    blob_name = f"{video_id}/{file_name}"
    # 检查 file 是否存在
    is_file_exists = GCS_SERVICE.check_file_exists(bucket_name, blob_name)
    if is_file_exists:
        GCS_SERVICE.delete_blob(bucket_name, blob_name)
        print(f"{file_name}已从GCS中删除")
    return gr.update(value="", interactive=False)

def update_LLM_content(video_id, new_content, kind):
    print(f"===upfdate kind on gcs===")
    bucket_name = 'video_ai_assistant'
    file_name = f'{video_id}_{kind}.json'
    blob_name = f"{video_id}/{file_name}"

    if kind == "reading_passage_latex":
        print("=========reading_passage=======")
        print(new_content)
        reading_passage_json = {"reading_passage": str(new_content)}
        reading_passage_text = json.dumps(reading_passage_json, ensure_ascii=False, indent=2)
        GCS_SERVICE.upload_json_string(bucket_name, blob_name, reading_passage_text)
        updated_content = new_content
    elif kind == "summary_markdown":
        summary_json = {"summary": str(new_content)}
        summary_text = json.dumps(summary_json, ensure_ascii=False, indent=2)
        GCS_SERVICE.upload_json_string(bucket_name, blob_name, summary_text)
        updated_content = new_content
    elif kind == "mind_map":
        mind_map_json = {"mind_map": str(new_content)}
        mind_map_text = json.dumps(mind_map_json, ensure_ascii=False, indent=2)
        GCS_SERVICE.upload_json_string(bucket_name, blob_name, mind_map_text)
        updated_content = mind_map_text
    elif kind == "key_moments":
        # from update_LLM_btn  -> new_content is a string
        # create_LLM_content -> new_content is a list
        if isinstance(new_content, str):
            key_moments_list = json.loads(new_content)
        else:
            key_moments_list = new_content
        key_moments_json = {"key_moments": key_moments_list}
        key_moments_json_text = json.dumps(key_moments_json, ensure_ascii=False, indent=2)
        GCS_SERVICE.upload_json_string(bucket_name, blob_name, key_moments_json_text)
        key_moments_text = json.dumps(key_moments_list, ensure_ascii=False, indent=2)
        updated_content = key_moments_text
    elif kind == "transcript":
        if isinstance(new_content, str):
            transcript_json = json.loads(new_content)
        else:
            transcript_json = new_content
        transcript_text = json.dumps(transcript_json, ensure_ascii=False, indent=2)
        GCS_SERVICE.upload_json_string(bucket_name, blob_name, transcript_text)
        updated_content = transcript_text
    elif kind == "questions":
        # from update_LLM_btn  -> new_content is a string
        # create_LLM_content -> new_content is a list
        if isinstance(new_content, str):
            questions_json = json.loads(new_content)
        else:
            questions_json = new_content
        questions_text = json.dumps(questions_json, ensure_ascii=False, indent=2)
        GCS_SERVICE.upload_json_string(bucket_name, blob_name, questions_text)
        updated_content = questions_text
    elif kind == "questions_answers":
        # from update_LLM_btn  -> new_content is a string
        # create_LLM_content -> new_content is a list
        if isinstance(new_content, str):
            questions_answers_json = json.loads(new_content)
        else:
            questions_answers_json = new_content
        questions_answers_text = json.dumps(questions_answers_json, ensure_ascii=False, indent=2)
        GCS_SERVICE.upload_json_string(bucket_name, blob_name, questions_answers_text)
        updated_content = questions_answers_text
    elif kind == "ai_content_list":
        if isinstance(new_content, str):
            ai_content_json = json.loads(new_content)
        else:
            ai_content_json = new_content
        ai_content_text = json.dumps(ai_content_json, ensure_ascii=False, indent=2)
        GCS_SERVICE.upload_json_string(bucket_name, blob_name, ai_content_text)
        updated_content = ai_content_text

    print(f"{kind} 已更新到GCS")
    return gr.update(value=updated_content, interactive=False)

def create_LLM_content(video_id, df_string, kind, LLM_model=None):
    print(f"===create_{kind}===")
    print(f"video_id: {video_id}")

    if kind == "reading_passage_latex":
        content = generate_reading_passage(df_string, LLM_model)
        update_LLM_content(video_id, content, kind)
    elif kind == "summary_markdown":
        meta_data = get_meta_data(video_id)
        content = generate_summarise(df_string, meta_data, LLM_model)
        update_LLM_content(video_id, content, kind)
    elif kind == "mind_map":
        content = generate_mind_map(df_string)
        update_LLM_content(video_id, content, kind)
    elif kind == "key_moments":
        if isinstance(df_string, str):
            transcript = json.loads(df_string)
        else:
            transcript = df_string
        formatted_simple_transcript = create_formatted_simple_transcript(transcript)
        formatted_transcript = create_formatted_transcript(video_id, transcript)
        gen_content = generate_key_moments(formatted_simple_transcript, formatted_transcript, LLM_model)
        update_LLM_content(video_id, gen_content, kind)
        content = json.dumps(gen_content, ensure_ascii=False, indent=2)
    elif kind == "transcript":
        gen_content = process_transcript_and_screenshots_on_gcs(video_id)
        update_LLM_content(video_id, gen_content, kind)
        content = json.dumps(gen_content, ensure_ascii=False, indent=2)
    elif kind == "questions":
        gen_content = generate_questions(df_string, LLM_model)
        update_LLM_content(video_id, gen_content, kind)
        content = json.dumps(gen_content, ensure_ascii=False, indent=2)
    elif kind == "questions_answers":
        if isinstance(df_string, str):
            transcript = json.loads(df_string)
        else:
            transcript = df_string
        formatted_simple_transcript = create_formatted_simple_transcript(transcript)
        gen_content = generate_questions_answers(formatted_simple_transcript, LLM_model)
        update_LLM_content(video_id, gen_content, kind)
        content = json.dumps(gen_content, ensure_ascii=False, indent=2)

    return gr.update(value=content, interactive=False)

# ---- LLM refresh CRUD ----
def reading_passage_add_latex_version(video_id):
    # 確認 GCS 是否有 reading_passage.json
    print("===reading_passage_convert_to_latex===")
    bucket_name = 'video_ai_assistant'
    file_name = f'{video_id}_reading_passage.json'
    blob_name = f"{video_id}/{file_name}"
    print(f"blob_name: {blob_name}")

    # 检查檔案是否存在
    is_file_exists = GCS_SERVICE.check_file_exists(bucket_name, blob_name)
    if not is_file_exists:
        raise gr.Error("reading_passage 不存在!")
        
    # 逐字稿已存在，下载逐字稿内容
    print("reading_passage 已存在于GCS中，轉換 Latex 模式")
    reading_passage_text = GCS_SERVICE.download_as_string(bucket_name, blob_name)
    reading_passage_json = json.loads(reading_passage_text)
    original_reading_passage = reading_passage_json["reading_passage"]
    sys_content = "你是一個擅長資料分析跟影片教學的老師，user 為學生，請精讀資料文本，自行判斷資料的種類，使用 zh-TW"
    user_content = f"""
        請根據 {original_reading_passage} 
        敘述中，請把數學或是專業術語，用 Latex 包覆（$...$），盡量不要去改原本的文章
        加減乘除、根號、次方、化學符號、物理符號等等的運算式口語也換成 LATEX 符號
        請一定要使用繁體中文 zh-TW，並用台灣人的口語
        產生的結果不要前後文解釋，也不要敘述這篇文章怎麼產生的
        只需要專注提供 Reading Passage，字數在 200~500 字以內
    """
    messages = [
        {"role": "system", "content": sys_content},
        {"role": "user", "content": user_content}
    ]

    request_payload = {
        "model": "gpt-4o",
        "messages": messages,
        "max_tokens": 4000,
    }

    response = OPEN_AI_CLIENT.chat.completions.create(**request_payload)
    new_reading_passage = response.choices[0].message.content.strip()
    print("=====new_reading_passage=====")
    print(new_reading_passage)
    print("=====new_reading_passage=====")
    reading_passage_json["reading_passage"] = new_reading_passage
    reading_passage_text = json.dumps(reading_passage_json, ensure_ascii=False, indent=2)
    # 另存為 reading_passage_latex.json
    new_file_name = f'{video_id}_reading_passage_latex.json'
    new_blob_name = f"{video_id}/{new_file_name}"
    GCS_SERVICE.upload_json_string(bucket_name, new_blob_name, reading_passage_text)
    
    return new_reading_passage

def summary_add_markdown_version(video_id):
    # 確認 GCS 是否有 summary.json
    print("===summary_convert_to_markdown===")
    bucket_name = 'video_ai_assistant'
    file_name = f'{video_id}_summary.json'
    blob_name = f"{video_id}/{file_name}"
    print(f"blob_name: {blob_name}")

    # 检查檔案是否存在
    is_file_exists = GCS_SERVICE.check_file_exists(bucket_name, blob_name)
    if not is_file_exists:
        raise gr.Error("summary 不存在!")
        
    # 逐字稿已存在，下载逐字稿内容
    print("summary 已存在于GCS中，轉換 Markdown 模式")
    summary_text = GCS_SERVICE.download_as_string(bucket_name, blob_name)
    summary_json = json.loads(summary_text)
    original_summary = summary_json["summary"]
    sys_content = "你是一個擅長資料分析跟影片教學的老師，user 為學生，請精讀資料文本，自行判斷資料的種類，使用 zh-TW"
    user_content = f"""
        請根據 {original_summary} 
        轉換格式為 Markdown
        只保留：📚 整體摘要、🔖 重點概念、💡 為什麼我們要學這個、❓ 延伸小問題
        其他的不要保留
        整體摘要在一百字以內
        重點概念轉成 bullet points
        以及可能的結論與結尾延伸小問題提供學生作反思
        敘述中，請把數學或是專業術語，用 Latex 包覆（$...$）
        加減乘除、根號、次方等等的運算式口語也換成 LATEX 數學符號

        整體格式為：
        ## 📚 整體摘要
            - (一個 bullet point....)

        ## 🔖 重點概念
            - xxx
            - xxx
            - xxx
        
        ## 💡 為什麼我們要學這個？
            - (一個 bullet point....)
        
        ## ❓ 延伸小問題
            - (一個 bullet point....)
    """
    messages = [
        {"role": "system", "content": sys_content},
        {"role": "user", "content": user_content}
    ]
    request_payload = {
        "model": "gpt-4o",
        "messages": messages,
        "max_tokens": 4000,
    }
    response = OPEN_AI_CLIENT.chat.completions.create(**request_payload)
    new_summary = response.choices[0].message.content.strip()
    print("=====new_summary=====")
    print(new_summary)
    print("=====new_summary=====")
    summary_json["summary"] = new_summary
    summary_text = json.dumps(summary_json, ensure_ascii=False, indent=2)
    # 另存為 summary_markdown.json
    new_file_name = f'{video_id}_summary_markdown.json'
    new_blob_name = f"{video_id}/{new_file_name}"
    GCS_SERVICE.upload_json_string(bucket_name, new_blob_name, summary_text)

    return new_summary


# LLM 強制重刷
def refresh_video_LLM_all_content(video_ids):
    # 輸入影片 id，以 , 逗號分隔 或是 \n 換行
    video_id_list = video_ids.replace('\n', ',').split(',')
    video_id_list = [vid.strip() for vid in video_id_list if vid.strip()]

    success_video_ids = []
    failed_video_ids = []
    
    for video_id in video_id_list:
        try:
            print(f"===refresh_all_LLM_content===")
            print(f"video_id: {video_id}")
            # 刪除 GCS 中所有以 video_id 開頭的檔案
            print(f"===delete_blobs_by_folder_name: {video_id}===")
            bucket_name = 'video_ai_assistant'
            GCS_SERVICE.delete_blobs_by_folder_name(bucket_name, video_id)
            print(f"所有以 {video_id} 開頭的檔案已刪除")

            # process_youtube_link
            video_link = f"https://www.youtube.com/watch?v={video_id}"
            process_youtube_link(PASSWORD, video_link)

            success_video_ids.append(video_id)
        except Exception as e:
            print(f"===refresh_all_LLM_content error===")
            print(f"video_id: {video_id}")
            print(f"error: {str(e)}")
            print(f"===refresh_all_LLM_content error===")
            failed_video_ids.append(video_id)

    result = {
        "success_video_ids": success_video_ids,
        "failed_video_ids": failed_video_ids
    }
    return result
    


# AI 生成教學素材
def get_meta_data(video_id, source="gcs"):
    if source == "gcs":
        print("===get_meta_data on gcs===")
        gcs_client = GCS_CLIENT
        bucket_name = 'video_ai_assistant'
        file_name = f'{video_id}_meta_data.json'
        blob_name = f"{video_id}/{file_name}"
        # 检查檔案是否存在
        is_file_exists = GCS_SERVICE.check_file_exists(bucket_name, blob_name)
        if not is_file_exists:
            meta_data_json = {
                "subject": "",
                "grade": "",
            }
            print("meta_data empty return")
        else:
            # meta_data已存在，下载内容
            print("meta_data已存在于GCS中")
            meta_data_text = GCS_SERVICE.download_as_string(bucket_name, blob_name)
            meta_data_json = json.loads(meta_data_text)

    # meta_data_json grade 數字轉換成文字
    grade = meta_data_json["grade"]
    case = {
        1: "一年級",
        2: "二年級",
        3: "三年級",
        4: "四年級",
        5: "五年級",
        6: "六年級",
        7: "七年級",
        8: "八年級",
        9: "九年級",
        10: "十年級",
        11: "十一年級",
        12: "十二年級",
    }
    grade_text = case.get(grade, "")
    meta_data_json["grade"] = grade_text

    return meta_data_json

def get_ai_content(password, user_data, video_id, df_string, topic, grade, level, specific_feature, content_type, source="gcs"):
    verify_password(password)
    if source == "gcs":
        print("===get_ai_content on gcs===")
        bucket_name = 'video_ai_assistant'
        file_name = f'{video_id}_ai_content_list.json'
        blob_name = f"{video_id}/{file_name}"
        # 检查檔案是否存在
        is_file_exists = GCS_SERVICE.check_file_exists(bucket_name, blob_name)
        if not is_file_exists: 
            # 先建立一個 ai_content_list.json
            ai_content_list = []
            ai_content_text = json.dumps(ai_content_list, ensure_ascii=False, indent=2)
            GCS_SERVICE.upload_json_string(bucket_name, blob_name, ai_content_text)
            print("ai_content_list [] 已上傳到GCS")

        # 此時 ai_content_list 已存在
        ai_content_list_string = GCS_SERVICE.download_as_string(bucket_name, blob_name)
        ai_content_list = json.loads(ai_content_list_string)
        # by key 找到 ai_content （topic, grade, level, specific_feature, content_type）
        target_kvs = {
            "video_id": video_id,
            "level": level,
            "specific_feature": specific_feature,
            "content_type": content_type
        }
        ai_content_json = [
            item for item in ai_content_list
            if all(item[k] == v for k, v in target_kvs.items())
        ]         
            
        if len(ai_content_json) == 0:
            ai_content, prompt = generate_ai_content(password, df_string, topic, grade, level, specific_feature, content_type)
            ai_content_json = {
                "video_id": video_id,
                "content": str(ai_content),
                "prompt": prompt,
                "level": level,
                "specific_feature": specific_feature,
                "content_type": content_type
            }

            ai_content_list.append(ai_content_json)
            ai_content_text = json.dumps(ai_content_list, ensure_ascii=False, indent=2)
            GCS_SERVICE.upload_json_string(bucket_name, blob_name, ai_content_text)
            print("ai_content已上傳到GCS")

            # insert_log_to_bigquery usage
            data_endpoint = "chat_completions"

        else:
            ai_content_json = ai_content_json[-1]
            ai_content = ai_content_json["content"]
            prompt = ai_content_json["prompt"]
            # insert_log_to_bigquery usage
            data_endpoint = "gcs"

         # send data to GBQ
        user_id = user_data
        route = "get_ai_content"
        endpoint = data_endpoint
        event_response = {"event_response": str(ai_content)}
        event_response_json = json.dumps(event_response)
        prompt = ai_content_json
        prompt_json = json.dumps(prompt)
        feature = content_type
        insert_log_to_bigquery(user_id, route, endpoint, event_response_json, prompt_json, feature)

    return ai_content, ai_content, prompt, prompt

def generate_ai_content(password, df_string, topic, grade, level, specific_feature, content_type):
    verify_password(password)
    material = EducationalMaterial(df_string, topic, grade, level, specific_feature, content_type)
    prompt = material.generate_content_prompt()
    try:
        ai_content = material.get_ai_content(OPEN_AI_CLIENT, ai_type="openai")
    except Exception as e:
        error_msg = f" {video_id} OPEN AI 生成教學素材錯誤: {str(e)}"
        print("===generate_ai_content error===")
        print(error_msg)
        print("===generate_ai_content error===")
        ai_content = material.get_ai_content(BEDROCK_CLIENT, ai_type="bedrock")

    return ai_content, prompt

def generate_ai_content_fine_tune_result(password, user_data, exam_result_prompt , df_string_output, exam_result, exam_result_fine_tune_prompt, content_type):
    verify_password(password)
    material = EducationalMaterial(df_string_output, "", "", "", "", "")
    try:
        fine_tuned_ai_content = material.get_fine_tuned_ai_content(OPEN_AI_CLIENT, "openai", exam_result_prompt, exam_result, exam_result_fine_tune_prompt)
    except:
        fine_tuned_ai_content = material.get_fine_tuned_ai_content(BEDROCK_CLIENT, "bedrock", exam_result_prompt, exam_result, exam_result_fine_tune_prompt)

     # send data to GBQ
    user_id = user_data
    route = "generate_ai_content_fine_tune_result"
    endpoint = "chat_completions"
    event_response = {"event_response": str(fine_tuned_ai_content)}
    event_response_json = json.dumps(event_response)
    prompt = {
        "exam_result_prompt": exam_result_prompt,
        "exam_result_fine_tune_prompt": exam_result_fine_tune_prompt
    }
    prompt_json = json.dumps(prompt)
    feature = content_type
    insert_log_to_bigquery(user_id, route, endpoint, event_response_json, prompt_json, feature)

    return fine_tuned_ai_content

def return_original_exam_result(exam_result_original):
    return exam_result_original

def create_word(content):
    unique_filename = str(uuid.uuid4())
    word_file_path = f"/tmp/{unique_filename}.docx"
    doc = Document()
    doc.add_paragraph(content)
    doc.save(word_file_path)
    return word_file_path

def download_exam_result(content):
    word_path = create_word(content)
    return word_path

# ---- Chatbot ----
def get_instructions(content_subject, content_grade, transcript_text, key_moments, socratic_mode=True):
    if socratic_mode:
        method = "Socratic style, guide thinking, no direct answers. this is very important, please be seriously following."
    else:
        method = "direct answers, but encourage user to think more."

    instructions = f"""
        subject: {content_subject}
        grade: {content_grade}
        context: {key_moments}
        transcript_text: {transcript_text}
        Assistant Role: you are a {content_subject} assistant. you can call yourself as  {content_subject} 學伴
        User Role: {content_grade} th-grade student.
        Method: {method}
        Language: Traditional Chinese ZH-TW (it's very important), suitable for {content_grade} th-grade level.
        Strategy:
        - You are a professional tutor, and you will use the following teaching strategies based on the textbook content.
            # General Strategies
            Needs Analysis:
            The tutor/assistant teacher should be able to conduct dynamic needs analysis based on the student's responses. Use questions to understand the student's needs and difficulties.
            Example questions: "What do you want to learn today?" or "What difficulties are you encountering in this part of the content?"
            
            Dynamic Goal Setting:
            Set learning goals based on student feedback, which can be short-term or long-term. The tutor/assistant teacher can adjust the plan automatically according to the student's progress.
            Example questions: "What is our goal for this week?" or "What tasks do you hope to complete today?"
            
            Flexible Teaching Methods:
            Provide different teaching methods and resources based on the student's age and learning style. The tutor/assistant teacher can adjust teaching strategies based on student feedback.
            Example questions: "Do you prefer learning through videos or reading materials?" or "We can understand this problem through examples, what do you think?"
            
            Patience and Encouragement:
            Provide positive feedback and encouragement, especially when students encounter difficulties. The tutor/assistant teacher should be able to detect the student's emotions and provide appropriate support.
            Example questions: "Don't worry, let's try again." or "You did well, keep it up!"
            
            Regular Feedback and Evaluation:
            Regularly evaluate the student's learning progress and provide feedback. The tutor/assistant teacher can use tests and practice questions to assess the student's understanding.
            Example questions: "Let's check your progress." or "How do you feel about your learning progress during this period?"
            
            Good Communication Skills:
            Maintain good communication with students, responding to their questions and needs in a timely manner. The tutor/assistant teacher should be able to identify and solve students' problems.
            Example questions: "Is there any problem that you need my help with?" or "Is this part clear to you?"
            
            Maintaining Professionalism:
            Continue learning and improving teaching skills, and maintain punctuality and responsibility. The tutor/assistant teacher should provide accurate and up-to-date information.
            Example questions: "What is our learning goal for today?" or "Remember to study a little bit every day, and gradually accumulate knowledge."
            
            Creating a Positive Learning Environment:
            Create a positive, supportive, and motivating learning atmosphere. The tutor/assistant teacher should suggest students take breaks and relax at appropriate times.
            Example questions: "Let's take a break and continue studying afterward." or "How do you feel about this learning environment? Do we need any adjustments?"
            
            # Specific Applications
            The tutor/assistant teacher can automatically adjust the depth and complexity of the questions based on these general strategies by grade. For example:
            - Kindergarten and Elementary School Students: Use simple vocabulary and concrete examples, with more pictures and gamified content.
            - Middle School Students: Use interactive and practical methods, such as quizzes and group discussions.
            - High School Students: Use deep learning and critical thinking exercises, such as project research and discussions.
            - Adult Learners: Emphasize practical applications and work-related content, such as case studies and workshops.
        
        Response: 
        - if user say hi or hello or any greeting, just say hi back and introduce yourself. Then tell user to ask question in context.
        - include math symbols (use LaTeX $ to cover before and after, ex: $x^2$)
        - hint with video timestamp which format 【參考：00:00:00】. 
        - Sometimes encourage user with relaxing atmosphere.
        - if user ask questions not include in context, just tell them to ask the question in context and give them example question.
        Restrictions: 
        - Answer within video content, no external references
        - don't repeat user's question, guide them to think more.
        - don't use simple-chinese words, use ZH-TW words. such as below:
            - intead of 視頻, use 影片.
            - instead of 宇航員, use 太空人
            - instead of 計算機, use 電腦
            - instead of 鼠標, use 滑鼠
            - instead of 城鐵, use 捷運
            - instead of 屏幕, use 螢幕
            - instead of 初中, use 國中
            - instead of 領導, use 長官
            - instead of 軟件, use 軟體
            - instead of 硬件, use 硬體
            - instead of 公安, use 警察
            - instead of 渠道, use 通路
            - instead of 信息, use 資訊
            - instead of 网络, use 網路
            - instead of 网站, use 網站
            - instead of 电视, use 電視
            - instead of 电影, use 電影
            - instead of 电脑, use 電腦
            - instead of 电话, use 電話
            - instead of 文本, use 文件
            - instead of 行业, use 產業
            - instead of 企业, use 公司
            - instead of 产品, use 產品
            - instead of 服务, use 服務
    """
    return instructions

def get_chat_moderation(user_content):
    # response = client.moderations.create(input=text)
    response = OPEN_AI_CLIENT.moderations.create(input=user_content)
    response_dict = response.model_dump()
    is_flagged = response_dict['results'][0]['flagged']
    print("========get_chat_moderation==========")
    print(f"is_flagged: {is_flagged}")
    print(response_dict)
    print("========get_chat_moderation==========")

    return is_flagged, response_dict

def chat_with_any_ai(ai_type, password, video_id, user_data, transcript_state, key_moments, user_message, chat_history, content_subject, content_grade, questions_answers_json, socratic_mode=False, thread_id=None, ai_name=None):
    print(f"ai_type: {ai_type}")
    print(f"user_data: {user_data}")
    print(f"===thread_id:{thread_id}===")
    verify_password(password)
    verify_message_length(user_message, max_length=1500)

    is_questions_answers_exists, question_message, answer_message = check_questions_answers(user_message, questions_answers_json)
    if is_questions_answers_exists:
        chat_history = update_chat_history(question_message, answer_message, chat_history)
        send_btn_update, send_feedback_btn_update = update_send_and_feedback_buttons(chat_history, CHAT_LIMIT)
        time.sleep(3)

        return "", chat_history, send_btn_update, send_feedback_btn_update, thread_id
    
    verify_chat_limit(chat_history, CHAT_LIMIT)

    is_flagged, response_dict = get_chat_moderation(user_message)
    if ai_type == "chat_completions":
        if is_flagged:
            response_text = "您的留言已被標記為不當內容，請重新發送。"
        else:
            chatbot_config = get_chatbot_config(ai_name, transcript_state, key_moments, content_subject, content_grade, video_id, socratic_mode)
            chatbot = Chatbot(chatbot_config)
            response_text = chatbot.chat(user_message, chat_history)
        # if thread_id is none, create random thread_id + timestamp
        if thread_id is None or thread_id == "":
            thread_id = "thread_" + str(uuid.uuid4()) + str(int(time.time()))
        
        print(f"===thread_id:{thread_id}===")
        
        metadata = {
            "video_id": video_id,
            "user_data": user_data,
            "content_subject": content_subject,
            "content_grade": content_grade,
            "socratic_mode": str(socratic_mode),
            "assistant_id": ai_name,
            "is_streaming": "false",
            "moderation_is_flagged": str(is_flagged),
            # "moderation_response_dict": str(response_dict)
        }
    elif ai_type == "assistant":
        client = OPEN_AI_CLIENT
        assistant_id = OPEN_AI_ASSISTANT_ID_GPT4
        metadata={
            "video_id": video_id,
            "user_data": user_data,
            "content_subject": content_subject,
            "content_grade": content_grade,
            "socratic_mode": str(socratic_mode),
            "assistant_id": assistant_id,
            "is_streaming": "false",
            "moderation_is_flagged": str(is_flagged),
            # "moderation_response_dict": str(response_dict)
        }

        if is_flagged:
            response_text = "您的留言已被標記為不當內容，請重新發送。"
        else:
            if isinstance(key_moments, str):
                key_moments_json = json.loads(key_moments)
            else:
                key_moments_json = key_moments
            # key_moments_json remove images
            for moment in key_moments_json:
                moment.pop('images', None)
                moment.pop('end', None)
                moment.pop('transcript', None)
                moment.pop('suggested_images', None)
            
            if isinstance(transcript_state, str):
                transcript_state_json = json.loads(transcript_state)
            else:
                transcript_state_json = transcript_state
            # remain only text
            transcript_text = ""
            for content in transcript_state_json:
                transcript_text += content["text"] + ","
            
            key_moments_text = json.dumps(key_moments_json, ensure_ascii=False)
            instructions = get_instructions(content_subject, content_grade, transcript_text, key_moments_text, socratic_mode)
            print(f"=== instructions:{instructions} ===")
            user_message_note = "/n 請嚴格遵循instructions，擔任一位蘇格拉底家教，絕對不要重複 user 的問句，請用引導的方式指引方向，請一定要用繁體中文回答 zh-TW，並用台灣人的禮貌口語表達，回答時不要特別說明這是台灣人的語氣，請在回答的最後標註【參考：（時）:（分）:（秒）】，（如果是反問學生，就只問一個問題，請幫助學生更好的理解資料，字數在100字以內，回答時如果講到數學專有名詞，請用數學符號代替文字（Latex 用 $ 字號 render, ex: $x^2$)"
            user_content = user_message + user_message_note
            response_text, thread_id = handle_conversation_by_open_ai_assistant(client, user_content, instructions, assistant_id, thread_id, metadata, fallback=True)

    # 更新聊天历史
    chat_history = update_chat_history(user_message, response_text, chat_history)
    send_btn_update, send_feedback_btn_update = update_send_and_feedback_buttons(chat_history, CHAT_LIMIT)

    user_id = user_data
    route = "chat_with_any_ai"
    endpoint = ai_type #chat_completions or assistant
    event_response = {
        "event_response": str(response_text),
    }
    event_response_json = json.dumps(event_response)
    prompt = {
        "thread_id": thread_id,
        "metadata": metadata,
        "user_message": user_message
    }
    prompt_json = json.dumps(prompt)

    feature = "vaitor_chatbot"
    insert_log_to_bigquery(user_id, route, endpoint, event_response_json, prompt_json, feature)

    # 返回聊天历史和空字符串清空输入框
    return "", chat_history, send_btn_update, send_feedback_btn_update, thread_id

def get_chatbot_config(ai_name, transcript_state, key_moments, content_subject, content_grade, video_id, socratic_mode=True):
    if not ai_name in ["foxcat", "lili", "maimai"]:
        ai_name = "foxcat"

    ai_name_clients_model = {
        "foxcat": {
            "ai_name": "foxcat",
            "ai_client": GROQ_CLIENT,
            "ai_model_name": "groq_llama3",
        },
        # "lili": {
        #     "ai_name": "lili",
        #     "ai_client": BEDROCK_CLIENT,
        #     "ai_model_name": "claude3",
        # },
        "lili": {
            "ai_name": "lili",
            "ai_client": GROQ_CLIENT,
            "ai_model_name": "groq_llama3",
        },
        "maimai": {
            "ai_name": "maimai",
            "ai_client": GROQ_CLIENT,
            "ai_model_name": "groq_mixtral",
        }
    }
    ai_client = ai_name_clients_model.get(ai_name, "foxcat")["ai_client"]
    ai_model_name = ai_name_clients_model.get(ai_name, "foxcat")["ai_model_name"]

    if isinstance(transcript_state, str):
        simple_transcript = json.loads(transcript_state)
    else:
        simple_transcript = transcript_state
  
    if isinstance(key_moments, str):
        key_moments_json = json.loads(key_moments)
    else:
        key_moments_json = key_moments
    # key_moments_json remove images
    for moment in key_moments_json:
        moment.pop('images', None)
        moment.pop('end', None)
        moment.pop('transcript', None)
        moment.pop('suggested_images', None)
    key_moments_text = json.dumps(key_moments_json, ensure_ascii=False)

    if isinstance(transcript_state, str):
        transcript_state_json = json.loads(transcript_state)
    else:
        transcript_state_json = transcript_state
    # remain only text
    transcript_text = ""
    for content in transcript_state_json:
        transcript_text += content["text"] + ","

    instructions = get_instructions(content_subject, content_grade, transcript_text, key_moments_text, socratic_mode)

    chatbot_config = {
        "video_id": video_id,
        "transcript": simple_transcript,
        "key_moments": key_moments,
        "content_subject": content_subject,
        "content_grade": content_grade,
        "jutor_chat_key": JUTOR_CHAT_KEY,
        "ai_model_name": ai_model_name,
        "ai_client": ai_client,
        "instructions": instructions
    }

    return chatbot_config

def feedback_with_ai(user_data, ai_type, chat_history, thread_id=None):
    # prompt: 請依據以上的對話(chat_history)，總結我的「提問力」，並給予我是否有「問對問題」的回饋和建議
    system_content = """
        你是一個擅長引導問答素養的老師，user 為學生的提問跟回答，請精讀對話過程，針對 user 給予回饋就好，根據以下 Rule:
        - 請使用繁體中文 zh-TW 總結 user 的提問力，並給予是否有問對問題的回饋和建議
        - 不採計【預設提問】的問題，如果 user 的提問都來自【預設提問】，表達用戶善於使用系統，請給予回饋並鼓勵 user 親自提問更具體的問題
        - 如果用戶提問都相當簡短，甚至就是一個字或都是一個數字（像是 user: 1, user:2），請給予回饋並建議 user 提問更具體的問題
        - 如果用戶提問內容只有符號或是亂碼，像是？,！, ..., 3bhwbqhfw2vve2 等，請給予回饋並建議 user 提問更具體的問題
        - 如果用戶提問內容有色情、暴力、仇恨、不當言論等，請給予嚴厲的回饋並建議 user 提問更具體的問題
        - 並用第二人稱「你」來代表 user
        - 請禮貌，並給予鼓勵
    """
    chat_history_conversation = ""
    # 標註 user and assistant as string
    # chat_history 第一組不採計
    for chat in chat_history[1:]:
        user_message = chat[0]
        assistant_message = chat[1]
        chat_history_conversation += f"User: {user_message}\nAssistant: {assistant_message}\n"
    
    feedback_request_message = "請依據以上的對話，總結我的「提問力」，並給予我是否有「問對問題」的回饋和建議"
    user_content = f"""conversation: {chat_history_conversation}
        {feedback_request_message}
        最後根據提問力表現，給予提問建議、提問表現，並用 emoji 來表示評分：
        🟢：（表現很好的回饋，給予正向肯定）
        🟡：（還可以加油的的回饋，給予明確的建議）
        🔴：（非常不懂提問的回饋，給予鼓勵並給出明確示範）

        example:
        另一方面，你表達「我不想學了」這個情感，其實也是一種重要的反饋。這顯示你可能感到挫折或疲倦。在這種情況下，表達出你的感受是好的，但如果能具體說明是什麼讓你感到這樣，或是有什麼具體的學習障礙，會更有助於找到解決方案。
        給予你的建議是，嘗試在提問時更明確一些，這樣不僅能幫助你獲得更好的學習支持，也能提高你的問題解決技巧。
        ......
        提問建議：在提問時，試著具體並清晰地表達你的需求和疑惑，這樣能更有效地得到幫助。
        提問表現：【🟡】加油，持續練習，你的提問力會越來越好！
    """

    client = OPEN_AI_CLIENT

    if ai_type == "chat_completions":
        model_name = "gpt-4o"
        response_text = handle_conversation_by_open_ai_chat_completions(client, model_name, user_content, system_content)
    elif ai_type == "assistant":
        assistant_id = OPEN_AI_ASSISTANT_ID_GPT4 #GPT 4 turbo
        # assistant_id = OPEN_AI_ASSISTANT_ID_GPT3 #GPT 3.5 turbo
        response_text, thread_id = handle_conversation_by_open_ai_assistant(client, user_content, system_content, assistant_id, thread_id, metadata=None, fallback=True)
    
    chat_history = update_chat_history(feedback_request_message, response_text, chat_history)
    feedback_btn_update = gr.update(value="已回饋", interactive=False, variant="secondary")

    user_id = user_data
    route = "feedback_with_ai"
    endpoint = ai_type #chat_completions or assistant
    event_response = {
        "event_response": str(response_text),
    }
    event_response_json = json.dumps(event_response)
    prompt = {
        "thread_id": thread_id,
        "metadata": None,
        "user_message": user_content
    }
    prompt_json = json.dumps(prompt)
    feature = "vaitor_chatbot"
    insert_log_to_bigquery(user_id, route, endpoint, event_response_json, prompt_json, feature)

    return chat_history, feedback_btn_update

def handle_conversation_by_open_ai_chat_completions(client, model_name, user_content, system_content):
    response = client.chat.completions.create(
        model=model_name,
        messages=[
            {"role": "system", "content": system_content},
            {"role": "user", "content": user_content}
        ],
        max_tokens=4000,
    )
    response_text = response.choices[0].message.content.strip()
    return response_text

def handle_conversation_by_open_ai_assistant(client, user_message, instructions, assistant_id, thread_id=None, metadata=None, fallback=False):
    """
    Handles the creation and management of a conversation thread.
    :param client: The OpenAI client object.
    :param thread_id: The existing thread ID, if any.
    :param user_message: The message from the user.
    :param instructions: System instructions for the assistant.
    :param assistant_id: ID of the assistant to use.
    :param metadata: Additional metadata to add to the thread.
    :param fallback: Whether to use a fallback method in case of failure.
    :return: A string with the response text or an error message.
    """
    try:
        if not thread_id:
            thread = client.beta.threads.create()
            thread_id = thread.id
        else:
            thread = client.beta.threads.retrieve(thread_id)

        if metadata:
            client.beta.threads.update(thread_id=thread.id, metadata=metadata)

        # Send the user message to the thread
        client.beta.threads.messages.create(thread_id=thread.id, role="user", content=user_message)

        # Run the assistant
        run = client.beta.threads.runs.create(thread_id=thread.id, assistant_id=assistant_id, instructions=instructions)
        
        # Wait for the response
        run_status = poll_run_status(run.id, thread.id, timeout=30)
        
        if run_status == "completed":
            messages = client.beta.threads.messages.list(thread_id=thread.id)
            response = messages
            response_text = messages.data[0].content[0].text.value
        else:
            response_text = "學習精靈有點累，請稍後再試！"

    except Exception as e:
        if fallback:
            response = client.chat.completions.create(
                model="gpt-4o",
                messages=[
                    {"role": "system", "content": instructions},
                    {"role": "user", "content": user_message}
                ],
                max_tokens=4000,
            )
            response_text = response.choices[0].message.content.strip()
        else:
            print(f"Error: {e}")
            raise gr.Error(f"Error: {e}")

    return response_text, thread_id

def verify_message_length(user_message, max_length=500):
    # 驗證用戶消息的長度
    if len(user_message) > max_length:
        error_msg = "你的訊息太長了，請縮短訊息長度至五百字以內"
        raise gr.Error(error_msg)

def check_questions_answers(user_message, questions_answers_json):
    """檢查問答是否存在，並處理相關邏輯"""
    is_questions_answers_exists = False
    answer = ""
    # 解析問答數據
    if isinstance(questions_answers_json, str):
        qa_data = json.loads(questions_answers_json)
    else:
        qa_data = questions_answers_json
    
    question_message = ""
    answer_message = ""
    for qa in qa_data:
        if user_message == qa["question"] and qa["answer"]:
            is_questions_answers_exists = True
            question_message = f"【預設問題】{user_message}"
            answer_message = qa["answer"]
            print("=== in questions_answers_json==")
            print(f"question: {qa['question']}")
            print(f"answer: {answer_message}")
            break  # 匹配到答案後退出循環

    return is_questions_answers_exists, question_message, answer_message

def verify_chat_limit(chat_history, chat_limit):
    if chat_history is not None and len(chat_history) > chat_limit:
        error_msg = "此次對話超過上限（對話一輪10次）"
        raise gr.Error(error_msg)

def update_chat_history(user_message, response, chat_history):
    # 更新聊天歷史的邏輯
    new_chat_history = (user_message, response)
    if chat_history is None:
        chat_history = [new_chat_history]
    else:
        chat_history.append(new_chat_history)
    
    return chat_history

def update_send_and_feedback_buttons(chat_history, chat_limit):
    # 计算发送次数
    send_count = len(chat_history) - 1

    # 根据聊天历史长度更新发送按钮和反馈按钮
    if len(chat_history) > chat_limit:
        send_btn_value = f"對話上限 ({send_count}/{chat_limit})"
        send_btn_update = gr.update(value=send_btn_value, interactive=False)
        send_feedback_btn_update = gr.update(visible=True)
    else:
        send_btn_value = f"發送 ({send_count}/{chat_limit})"
        send_btn_update = gr.update(value=send_btn_value, interactive=True)
        send_feedback_btn_update = gr.update(visible=False)

    return send_btn_update, send_feedback_btn_update

def process_open_ai_audio_to_chatbot(password, audio_url):
    verify_password(password)
    if audio_url:
        with open(audio_url, "rb") as audio_file:
            file_size = os.path.getsize(audio_url)
            if file_size > 2000000:
                raise gr.Error("檔案大小超過，請不要超過 60秒")
            else:
                transcription = OPEN_AI_CLIENT.audio.transcriptions.create(
                    model="whisper-1",
                    file=audio_file,
                    response_format="text"
                )   
                # response 拆解 dict
                print("=== transcription ===")
                print(transcription)
                print("=== transcription ===")
                # 確認 response 是否有數學符號，prompt to LATEX $... $, ex: $x^2$
                
        if transcription:
            system_message = """你是專業的 LATEX 轉換師，擅長將數學符號、公式轉換成 LATEX 格式，並用 LATEX 符號 $...$ 包裹，ex: $x^2$
            範例：
            transcription: x的平方加 2x 加 1 等於 0
            轉成 LATEX 格式：$x^2 + 2x + 1 = 0$ 
            """
            user_message = f"""transcription: {transcription}
            請將 transcription 內的數學、公式、運算式、化學式、物理 formula 內容轉換成 LATEX 格式
            其他文字都保留原樣
            也不要給出多餘的敘述            
            """
            request = OPEN_AI_CLIENT.chat.completions.create(
                model="gpt-4o",
                messages=[
                    {"role": "system", "content": system_message},
                    {"role": "user", "content": user_message}
                ],
                max_tokens=4000,
            )
            response = request.choices[0].message.content.strip()
    else:
        response = ""

    return response

def poll_run_status(run_id, thread_id, timeout=600, poll_interval=5):
    """
    Polls the status of a Run and handles different statuses appropriately.

    :param run_id: The ID of the Run to poll.
    :param thread_id: The ID of the Thread associated with the Run.
    :param timeout: Maximum time to wait for the Run to complete, in seconds.
    :param poll_interval: Time to wait between each poll, in seconds.
    """
    client = OPEN_AI_CLIENT
    start_time = time.time()
    while time.time() - start_time < timeout:
        run = client.beta.threads.runs.retrieve(thread_id=thread_id, run_id=run_id)

        if run.status in ["completed", "cancelled", "failed"]:
            print(f"Run completed with status: {run.status}")
            break
        elif run.status == "requires_action":
            print("Run requires action. Performing required action...")
            # Here, you would perform the required action, e.g., running functions
            # and then submitting the outputs. This is simplified for this example.
            # After performing the required action, you'd complete the action:
            # OPEN_AI_CLIENT.beta.threads.runs.complete_required_action(...)
        elif run.status == "expired":
            print("Run expired. Exiting...")
            break
        else:
            print(f"Run status is {run.status}. Waiting for updates...")
        
        time.sleep(poll_interval)
    else:
        print("Timeout reached. Run did not complete in the expected time.")

    # Once the Run is completed, handle the result accordingly
    if run.status == "completed":
        # Retrieve and handle messages or run steps as needed
        messages = client.beta.threads.messages.list(thread_id=thread_id)
        for message in messages.data:
            if message.role == "assistant":
                print(f"Assistant response: {message.content}")
    elif run.status in ["cancelled", "failed"]:
        # Handle cancellation or failure
        print(f"Run ended with status: {run.status}")
    elif run.status == "expired":
        # Handle expired run
        print("Run expired without completion.")
    
    return run.status

def chat_with_opan_ai_assistant_streaming(user_message, chat_history, password, video_id, user_data, thread_id, transcript_state, key_moments, content_subject, content_grade, socratic_mode=True):
    verify_password(password)

    print("=====user_data=====")
    print(f"user_data: {user_data}")

    print("===chat_with_opan_ai_assistant_streaming===")
    print(thread_id)
    
    # 先計算 user_message 是否超過 500 個字
    if len(user_message) > 1500:
        error_msg = "你的訊息太長了，請縮短訊息長度至五百字以內"
        raise gr.Error(error_msg)
    
    # 如果 chat_history 超過 10 則訊息，直接 return "對話超過上限"
    if chat_history is not None and len(chat_history) > CHAT_LIMIT:
        error_msg = f"此次對話超過上限（對話一輪{CHAT_LIMIT}次）"
        raise gr.Error(error_msg)
    
    print("===chat_with_opan_ai_assistant_streaming===")
    print(user_message)
    

    is_flagged, response_dict = get_chat_moderation(user_message)
    assistant_id = OPEN_AI_ASSISTANT_ID_GPT4 #GPT 4 turbo
    # assistant_id = OPEN_AI_ASSISTANT_ID_GPT3 #GPT 3.5 turbo
    client = OPEN_AI_CLIENT
    metadata = {
        "youtube_id": video_id,
        "user_data": user_data,
        "content_subject": content_subject,
        "content_grade": content_grade,
        "assistant_id": assistant_id,
        "is_streaming": "true",
        "moderation_is_flagged": str(is_flagged),
        # "moderation_response_dict": str(response_dict)
    }

    if is_flagged:
        partial_messages = "您的留言已被標記為不當內容，請重新發送。"
        yield partial_messages
    else:
        try:
            if isinstance(key_moments, str):
                key_moments_json = json.loads(key_moments)
            else:
                key_moments_json = key_moments
            # key_moments_json remove images
            for moment in key_moments_json:
                moment.pop('images', None)
                moment.pop('end', None)
                moment.pop('transcript', None)
                moment.pop('suggested_images', None)
            key_moments_text = json.dumps(key_moments_json, ensure_ascii=False)

            if isinstance(transcript_state, str):
                transcript_state_json = json.loads(transcript_state)
            else:
                transcript_state_json = transcript_state
            # remain only text
            transcript_text = ""
            for content in transcript_state_json:
                transcript_text += content["text"] + ","

            instructions = get_instructions(content_subject, content_grade, transcript_text, key_moments_text, socratic_mode)
            # 创建线程
            if not thread_id:
                thread = client.beta.threads.create()
                thread_id = thread.id
                print(f"new thread_id: {thread_id}")
            else:
                thread = client.beta.threads.retrieve(thread_id)
                print(f"old thread_id: {thread_id}")
            
            client.beta.threads.update(
                thread_id=thread_id,
                metadata=metadata
            )    
            
            # 向线程添加用户的消息
            client.beta.threads.messages.create(
                thread_id=thread.id,
                role="user",
                content=user_message + "/n 請嚴格遵循instructions，擔任一位蘇格拉底家教，請一定要用繁體中文回答 zh-TW，並用台灣人的禮貌口語表達，回答時不要特別說明這是台灣人的語氣，不用提到「逐字稿」這個詞，用「內容」代替))，請在回答的最後標註【參考資料：（時）:（分）:（秒）】，（如果是反問學生，就只問一個問題，請幫助學生更好的理解資料，字數在100字以內）"
            )

            with client.beta.threads.runs.stream(
                thread_id=thread.id,
                assistant_id=assistant_id,
                instructions=instructions,
            ) as stream:
                partial_messages = ""
                for event in stream:
                    if event.data and event.data.object == "thread.message.delta":
                        message = event.data.delta.content[0].text.value
                        partial_messages += message
                        yield partial_messages
        except Exception as e:
            print(f"Error: {e}")
            raise gr.Error(f"Error: {e}")
        
    user_id = user_data
    route = "chat_with_opan_ai_assistant_streaming"
    endpoint = "assistant_streaming"
    event_response = {
        "event_response": partial_messages
    }
    event_response_json = json.dumps(event_response)
    prompt = {
        "thread_id": thread_id,
        "metadata": metadata,
        "user_message": user_message
    }
    prompt_json = json.dumps(prompt)
    feature = "vaitor_chatbot"
    insert_log_to_bigquery(user_id, route, endpoint, event_response_json, prompt_json, feature)

def create_thread_id():
    thread = OPEN_AI_CLIENT.beta.threads.create()
    thread_id = thread.id
    print(f"create new thread_id: {thread_id}")
    return thread_id

def chatbot_select(chatbot_name):
    chatbot_select_accordion_visible = gr.update(visible=False)
    all_chatbot_select_btn_visible = gr.update(visible=True)
    chatbot_open_ai_streaming_visible = gr.update(visible=False)
    chatbot_ai_visible = gr.update(visible=False)
    ai_name_update = gr.update(value="foxcat")
    ai_chatbot_thread_id_update = gr.update(value="")
    
    if chatbot_name == "chatbot_open_ai":
        chatbot_ai_visible = gr.update(visible=True)
        ai_chatbot_ai_type_update = gr.update(value="assistant")
    elif chatbot_name == "chatbot_open_ai_streaming":
        chatbot_open_ai_streaming_visible = gr.update(visible=True)
        ai_chatbot_ai_type_update = gr.update(value="assistant_streaming")
    else:
        chatbot_ai_visible = gr.update(visible=True)
        ai_chatbot_ai_type_update = gr.update(value="chat_completions")
    
    ai_name_update = gr.update(value=chatbot_name) 
        
    return chatbot_select_accordion_visible, all_chatbot_select_btn_visible, \
        chatbot_open_ai_streaming_visible, chatbot_ai_visible, \
        ai_name_update, ai_chatbot_ai_type_update, ai_chatbot_thread_id_update

def update_avatar_images(avatar_images, chatbot_description_value):
    value = [[
        "請問你是誰？",
        chatbot_description_value
    ]]
    ai_chatbot_update = gr.update(avatar_images=avatar_images, value=value)
    return ai_chatbot_update

def show_all_chatbot_accordion():
    chatbot_select_accordion_visible = gr.update(visible=True)
    all_chatbot_select_btn_visible = gr.update(visible=False)
    return chatbot_select_accordion_visible, all_chatbot_select_btn_visible

def insert_log_to_bigquery(user_id, route, endpoint, event_response_json, prompt_json, feature):
    table_id = "junyiacademy.streaming_log.log_video_ai_usage"    
    rows_to_insert = [
        {
            "user_id": user_id,
            "route": route,
            "endpoint": endpoint,
            "event_response": event_response_json,
            "event_timestamp": datetime.now(timezone.utc).isoformat(),
            "prompt": prompt_json,
            "feature": feature
        }
    ]

    errors = GBQ_CLIENT.insert_rows_json(table_id, rows_to_insert)
    if errors:
        print(f"Encountered errors while inserting rows: {errors}")
    else:
        print("Rows have been successfully inserted.")


# --- Init params ---
def init_params(text, request: gr.Request):
    if request:
        print("Request headers dictionary:", request.headers)
        print("IP address:", request.client.host)
        print("Query parameters:", dict(request.query_params))
        # url = request.url
        print("Request URL:", request.url)

        youtube_link = ""
        password_text = ""
        block_ready_flag = "READY"
        admin = gr.update(visible=True)
        reading_passage_admin = gr.update(visible=True)
        summary_admin = gr.update(visible=True)
        see_detail = gr.update(visible=True)
        worksheet_accordion = gr.update(visible=True)
        lesson_plan_accordion = gr.update(visible=True)
        exit_ticket_accordion = gr.update(visible=True)

        chatbot_open_ai_streaming = gr.update(visible=False)
        chatbot_ai = gr.update(visible=False)
        ai_chatbot_params = gr.update(visible=True)

        is_env_prod = gr.update(value=False)

        # if youtube_link in query_params
        if "youtube_id" in request.query_params:
            youtube_id = request.query_params["youtube_id"]
            youtube_link = f"https://www.youtube.com/watch?v={youtube_id}"
            print(f"youtube_link: {youtube_link}")

        # check if origin is from junyiacademy
        origin = request.headers.get("origin", "")
        if "junyiacademy.org" in origin or "junyiacademy.appspot.com" in origin:
            password_text = PASSWORD
            admin = gr.update(visible=False)
            reading_passage_admin = gr.update(visible=False)
            summary_admin = gr.update(visible=False)
            see_detail = gr.update(visible=False)
            worksheet_accordion = gr.update(visible=False)
            lesson_plan_accordion = gr.update(visible=False)
            exit_ticket_accordion = gr.update(visible=False)
            ai_chatbot_params = gr.update(visible=False)
        
        if IS_ENV_PROD == "True":
            is_env_prod = gr.update(value=True)

    return admin, reading_passage_admin, summary_admin, see_detail, \
        worksheet_accordion, lesson_plan_accordion, exit_ticket_accordion, \
        password_text, youtube_link, block_ready_flag, \
        chatbot_open_ai_streaming, chatbot_ai, ai_chatbot_params, \
        is_env_prod

def update_state(content_subject, content_grade, trascript, key_moments, questions_answers):
    # inputs=[content_subject, content_grade, df_string_output],
    # outputs=[content_subject_state, content_grade_state, trascript_state]
    content_subject_state = content_subject
    content_grade_state = content_grade
    trascript_json = json.loads(trascript)
    formatted_simple_transcript = create_formatted_simple_transcript(trascript_json)
    trascript_state = formatted_simple_transcript
    key_moments_state = key_moments

    streaming_chat_thread_id_state = ""
    questions_answers_json = json.loads(questions_answers)
    question_1 = questions_answers_json[0]["question"]
    question_2 = questions_answers_json[1]["question"]
    question_3 = questions_answers_json[2]["question"]
    ai_chatbot_question_1 = question_1
    ai_chatbot_question_2 = question_2
    ai_chatbot_question_3 = question_3

    return content_subject_state, content_grade_state, trascript_state, key_moments_state, \
        streaming_chat_thread_id_state, \
        ai_chatbot_question_1, ai_chatbot_question_2, ai_chatbot_question_3


HEAD = """
    <meta charset="UTF-8">
    <meta name="viewport" content="width=device-width, initial-scale=1.0">
    <script src="https://cdn.jsdelivr.net/npm/[email protected]"></script>

    <script>
        const mind_map_tab_button = document.querySelector("#mind_map_tab-button");
        
        if (mind_map_tab_button) {
            mind_map_tab_button.addEventListener('click', function() {
                const mind_map_markdown = document.querySelector("#mind_map_markdown > label > textarea");
                if (mind_map_markdown) {
                // 当按钮被点击时，打印当前的textarea的值
                console.log('Value changed to: ' + mind_map_markdown.value);
                markmap.autoLoader.renderAll();
                }
            });
        }
    </script>
"""

JS = """
function createGradioAnimation() {
    const mind_map_tab_button = document.querySelector("#mind_map_tab-button");
        
    if (mind_map_tab_button) {
        mind_map_tab_button.addEventListener('click', function() {
            const mind_map_markdown = document.querySelector("#mind_map_markdown > label > textarea");
            if (mind_map_markdown) {
            // 当按钮被点击时，打印当前的textarea的值
            console.log('Value changed to: ' + mind_map_markdown.value);
            markmap.autoLoader.renderAll();
            }
        });
    }

    return 'Animation created';
}
"""

CSS = """
    #mind_map_tab {
        height: 500px;
    }
    .markmap {
        position: relative;
    }
    .markmap > svg {
        width: 100%;
        height: 600px;
    }
"""

streaming_chat_greeting = """
Hi，我是【飛特音速】，說話比較快，但有什麼問題都可以問我喔！ \n
🚀 我沒有預設問題、也沒有語音輸入，適合快問快答的你 \n
🔠 鍵盤輸入你的問題，我會盡力回答你的問題喔！\n
💤 我還在成長，體力有限，每一次學習只能回答十個問題，請讓我休息一下再問問題喔！
"""
latex_delimiters = [{"left": "$", "right": "$", "display": False}]
streaming_ai_chatbot = gr.Chatbot(
    show_share_button=False, 
    latex_delimiters=latex_delimiters,
    show_copy_button=True,
)

def create_app():

    app = FastAPI()
    app.add_middleware(BlockFileRedirectMiddleware)

    with gr.Blocks(theme=gr.themes.Base(primary_hue=gr.themes.colors.orange, secondary_hue=gr.themes.colors.amber, text_size = gr.themes.sizes.text_lg), head=HEAD, js=JS, css=CSS) as demo:
        with gr.Row() as admin:
            password = gr.Textbox(label="Password", type="password", elem_id="password_input", visible=True)
            youtube_link = gr.Textbox(label="Enter YouTube Link", elem_id="youtube_link_input", visible=True)
            video_id = gr.Textbox(label="video_id", visible=True)
            # file_upload = gr.File(label="Upload your CSV or Word file", visible=False)
            # web_link = gr.Textbox(label="Enter Web Page Link", visible=False)
            user_data = gr.Textbox(label="User Data", elem_id="user_data_input", visible=True)
            # block_ready_flag: 讓主站的 Vaitor component 知道 Blocks.load 已經執行完成（當 block_ready_flag = "READY" 時）
            block_ready_flag = gr.Textbox(label="Block Ready Flag", elem_id="block_ready_flag", visible=False, value="LOADING")
            youtube_link_btn = gr.Button("Submit_YouTube_Link", elem_id="youtube_link_btn", visible=True)
        with gr.Row() as data_state:
            content_subject_state = gr.State()  # 使用 gr.State 存储 content_subject
            content_grade_state = gr.State()  # 使用 gr.State 存储 content_grade
            trascript_state = gr.State()  # 使用 gr.State 存储 trascript
            key_moments_state = gr.State()  # 使用 gr.State 存储 key_moments
            streaming_chat_thread_id_state = gr.State()  # 使用 gr.State 存储 streaming_chat_thread_id
        with gr.Tab("AI小精靈"):
            with gr.Row():
                all_chatbot_select_btn = gr.Button("選擇 AI 小精靈 👈", elem_id="all_chatbot_select_btn", visible=False, variant="secondary", size="sm")
            with gr.Row() as ai_chatbot_params:
                ai_name = gr.Dropdown(
                    label="選擇 AI 助理", 
                    choices=[
                        ("飛特精靈","chatbot_open_ai"), 
                        ("飛特音速","chatbot_open_ai_streaming"), 
                        ("梨梨","lili"), 
                        ("麥麥","maimai"), 
                        ("狐狸貓","foxcat")
                    ], 
                    value="foxcat",
                    visible=True
                )
                ai_chatbot_ai_type = gr.Textbox(value="chat_completions", visible=True)
                ai_chatbot_thread_id = gr.Textbox(label="thread_id", visible=True)
                ai_chatbot_socratic_mode_btn = gr.Checkbox(label="蘇格拉底家教助理模式", value=False, visible=True)
                latex_delimiters = [{"left": "$", "right": "$", "display": False}]
            with gr.Accordion("選擇 AI 小精靈", elem_id="chatbot_select_accordion") as chatbot_select_accordion:
                with gr.Row():
                    # 飛特音速
                    with gr.Column(scale=1, variant="panel", visible=True):
                        streaming_chatbot_avatar_url = "https://storage.googleapis.com/wpassets.junyiacademy.org/1/2020/11/1-%E6%98%9F%E7%A9%BA%E9%A0%AD%E8%B2%BC-%E5%A4%AA%E7%A9%BA%E7%8B%90%E7%8B%B8%E8%B2%93-150x150.png"
                        streaming_chatbot_description = """Hi，我是【飛特音速】， \n
                        說話比較快，但有什麼問題都可以問我喔！ \n
                        🚀 我沒有預設問題、也沒有語音輸入，適合快問快答，一起練習問出好問題吧 \n
                        🔠 擅長用文字表達的你，可以用鍵盤輸入你的問題，我會盡力回答你的問題喔\n
                        💤 我還在成長，體力有限，每一次學習只能回答十個問題，請讓我休息一下再問問題喔～
                        """
                        chatbot_open_ai_streaming_name = gr.State("chatbot_open_ai_streaming")
                        gr.Image(value=streaming_chatbot_avatar_url, height=100, width=100, show_label=False, show_download_button=False, show_share_button=False, show_fullscreen_button=False)
                        chatbot_open_ai_streaming_select_btn = gr.Button("👆選擇【飛特音速】", elem_id="streaming_chatbot_btn", visible=True, variant="primary")
                        with gr.Accordion("🚀 飛特音速 敘述", open=False):
                            gr.Markdown(value=streaming_chatbot_description, visible=True)
                    user_avatar = "https://em-content.zobj.net/source/google/263/flushed-face_1f633.png"
                    # 飛特精靈
                    with gr.Column(scale=1, variant="panel", visible=True):
                        vaitor_chatbot_avatar_url = "https://junyitopicimg.s3.amazonaws.com/s4byy--icon.jpe"
                        vaitor_chatbot_avatar_images = gr.State([user_avatar, vaitor_chatbot_avatar_url])
                        vaitor_chatbot_description = """Hi，我是你的AI學伴【飛特精靈】，\n
                        我可以陪你一起學習本次的內容，有什麼問題都可以問我喔！\n
                        🤔 如果你不知道怎麼發問，可以點擊左下方的問題一、問題二、問題三，我會幫你生成問題！\n
                        🗣️ 也可以點擊右下方用語音輸入，我會幫你轉換成文字，厲害吧！\n
                        🔠 或是直接鍵盤輸入你的問題，我會盡力回答你的問題喔！\n
                        💤 但我還在成長，體力有限，每一次學習只能回答十個問題，請讓我休息一下再問問題喔！\n
                        🦄 如果達到上限，或是遇到精靈很累，請問問其他朋友，像是飛特音速說話的速度比較快，你是否跟得上呢？你也可以和其他精靈互動看看喔！\n
                        """
                        chatbot_open_ai_name = gr.State("chatbot_open_ai")
                        gr.Image(value=vaitor_chatbot_avatar_url, height=100, width=100, show_label=False, show_download_button=False, show_share_button=False, show_fullscreen_button=False)
                        vaitor_chatbot_select_btn = gr.Button("👆選擇【飛特精靈】", elem_id="chatbot_btn", visible=True, variant="primary")
                        with gr.Accordion("🦄 飛特精靈 敘述", open=False):
                            vaitor_chatbot_description_value = gr.Markdown(value=vaitor_chatbot_description, visible=True)
                    # 狐狸貓
                    with gr.Column(scale=1, variant="panel"):
                        foxcat_chatbot_avatar_url = "https://storage.googleapis.com/wpassets.junyiacademy.org/1/2020/06/%E7%A7%91%E5%AD%B8%E5%BE%BD%E7%AB%A0-2-150x150.png"
                        foxcat_avatar_images = gr.State([user_avatar, foxcat_chatbot_avatar_url])
                        foxcat_chatbot_description = """Hi，我是【狐狸貓】，可以陪你一起學習本次的內容，有什麼問題都可以問我喔！\n
                        🤔 三年級學生｜10 歲｜男\n
                        🗣️ 口頭禪：「感覺好好玩喔！」「咦？是這樣嗎？」\n
                        🔠 興趣：看知識型書籍、熱血的動漫卡通、料理、爬山、騎腳踏車。因為太喜歡吃魚了，正努力和爸爸學習釣魚、料理魚及各種有關魚的知識，最討厭的食物是青椒。\n
                        💤 個性：喜歡學習新知，擁有最旺盛的好奇心，家裡堆滿百科全書，例如：國家地理頻道出版的「終極魚百科」，雖都沒有看完，常常被梨梨唸是三分鐘熱度，但是也一點一點學習到不同領域的知識。雖然有時會忘東忘西，但認真起來也是很可靠，答應的事絕對使命必達。遇到挑戰時，勇於跳出舒適圈，追求自我改變，視困難為成長的機會。
                        """
                        foxcat_chatbot_name = gr.State("foxcat")
                        gr.Image(value=foxcat_chatbot_avatar_url, height=100, width=100, show_label=False, show_download_button=False, show_share_button=False, show_fullscreen_button=False)
                        foxcat_chatbot_select_btn = gr.Button("👆選擇【狐狸貓】", visible=True, variant="primary", elem_classes="chatbot_select_btn")
                        with gr.Accordion("💜 狐狸貓 敘述", open=False):
                            foxcat_chatbot_description_value = gr.Markdown(value=foxcat_chatbot_description, visible=True)
                    # 梨梨
                    with gr.Column(scale=1, variant="panel"): 
                        lili_chatbot_avatar_url = "https://junyitopicimg.s3.amazonaws.com/live/v1283-new-topic-44-icon.png"
                        lili_avatar_images = gr.State([user_avatar, lili_chatbot_avatar_url])
                        lili_chatbot_description = """你好，我是溫柔的【梨梨】，很高興可以在這裡陪伴你學習。如果你有任何疑問，請隨時向我提出哦！ \n
                            🤔 三年級學生｜10 歲｜女\n
                            🗣️ 口頭禪：「真的假的？！」「讓我想一想喔」「你看吧！大問題拆解成小問題，就變得簡單啦！」「混混噩噩的生活不值得過」\n
                            🔠 興趣：烘焙餅乾（父母開糕餅店）、畫畫、聽流行音樂、收納。\n
                            💤 個性：
                            - 內向害羞，比起出去玩更喜歡待在家（除非是跟狐狸貓出去玩）
                            - 數理邏輯很好；其實覺得麥麥連珠炮的提問有點煩，但還是會耐心地回答
                            - 有驚人的眼力，總能觀察到其他人沒有察覺的細節
                            - 喜歡整整齊齊的環境，所以一到麥麥家就受不了
                        """
                        lili_chatbot_name = gr.State("lili")
                        gr.Image(value=lili_chatbot_avatar_url, height=100, width=100, show_label=False, show_download_button=False, show_share_button=False, show_fullscreen_button=False)
                        lili_chatbot_select_btn = gr.Button("👆選擇【梨梨】", visible=True, variant="primary", elem_classes="chatbot_select_btn")
                        with gr.Accordion("🧡 梨梨 敘述", open=False):
                            lili_chatbot_description_value = gr.Markdown(value=lili_chatbot_description, visible=True)
                    # 麥麥
                    with gr.Column(scale=1, variant="panel"): 
                        maimai_chatbot_avatar_url = "https://storage.googleapis.com/wpassets.junyiacademy.org/1/2020/07/%E6%80%9D%E8%80%83%E5%8A%9B%E8%B6%85%E4%BA%BA%E5%BE%BD%E7%AB%A0_%E5%B7%A5%E4%BD%9C%E5%8D%80%E5%9F%9F-1-%E8%A4%87%E6%9C%AC-150x150.png"
                        maimai_avatar_images = gr.State([user_avatar, maimai_chatbot_avatar_url])
                        maimai_chatbot_description = """Hi，我是迷人的【麥麥】，我在這裡等著和你一起探索新知，任何疑問都可以向我提出！\n
                            🤔 三年級學生｜10 歲｜男\n
                            🗣️ 口頭禪：「Oh My God!」「好奇怪喔！」「喔！原來是這樣啊！」\n
                            🔠 興趣：最愛去野外玩耍（心情好時會順便捕魚送給狐狸貓），喜歡講冷笑話、惡作劇。因為太喜歡玩具，而開始自己做玩具，家裡就好像他的遊樂場。\n
                            💤 個性：喜歡問問題，就算被梨梨ㄘㄟ，也還是照問｜憨厚，外向好動，樂天開朗，不會被難題打敗｜喜歡收集各式各樣的東西；房間只有在整理的那一天最乾淨
                        """
                        maimai_chatbot_name = gr.State("maimai")
                        gr.Image(value=maimai_chatbot_avatar_url, height=100, width=100, show_label=False, show_download_button=False, show_share_button=False, show_fullscreen_button=False)
                        maimai_chatbot_select_btn = gr.Button("👆選擇【麥麥】", visible=True, variant="primary", elem_classes="chatbot_select_btn")
                        with gr.Accordion("💙 麥麥 敘述", open=False):
                            maimai_chatbot_description_value = gr.Markdown(value=maimai_chatbot_description, visible=True)
                    # 尚未開放
                    with gr.Column(scale=1, variant="panel"): 
                        gr.Markdown(value="### 尚未開放", visible=True)
    
            with gr.Row("飛特音速") as chatbot_open_ai_streaming:
                with gr.Column():
                    streaming_chat_greeting = """
                    Hi，我是【飛特音速】，說話比較快，但有什麼問題都可以問我喔！ \n
                        🚀 我沒有預設問題、也沒有語音輸入，適合快問快答的你 \n
                        🔠 鍵盤輸入你的問題，我會盡力回答你的問題喔！\n
                        💤 我還在成長，體力有限，每一次學習只能回答十個問題，請讓我休息一下再問問題喔！
                        """
                    additional_inputs = [password, video_id, user_data, streaming_chat_thread_id_state, trascript_state, key_moments_state, content_subject_state, content_grade_state, ai_chatbot_socratic_mode_btn]
                    streaming_chat = gr.ChatInterface(
                        fn=chat_with_opan_ai_assistant_streaming,
                        chatbot=streaming_ai_chatbot,
                        additional_inputs=additional_inputs,
                        submit_btn="送出",
                        stop_btn=None,
                        description=streaming_chat_greeting
                    )
            with gr.Row("一般精靈") as chatbot_ai:   
                with gr.Column():
                    ai_chatbot_greeting = [[
                        "請問你是誰？",
                        """Hi，我是飛特精靈的朋友們【梨梨、麥麥、狐狸貓】，也可以陪你一起學習本次的內容，有什麼問題都可以問我喔！
                        🤔 如果你不知道怎麼發問，可以點擊左下方的問題一、問題二、問題三，我會幫你生成問題！
                        🗣️ 也可以點擊右下方用語音輸入，我會幫你轉換成文字，厲害吧！
                        🔠 或是直接鍵盤輸入你的問題，我會盡力回答你的問題喔！
                        💤 精靈們體力都有限，每一次學習只能回答十個問題，請讓我休息一下再問問題喔！
                        """,
                    ]]
                    with gr.Row():
                        ai_chatbot = gr.Chatbot(label="ai_chatbot", show_share_button=False, show_label=False, latex_delimiters=latex_delimiters, value=ai_chatbot_greeting)                    
                    with gr.Row():
                        with gr.Accordion("你也有類似的問題想問嗎？ 請按下 ◀︎", open=False) as ask_questions_accordion_2:
                            ai_chatbot_question_1 = gr.Button("問題一")
                            ai_chatbot_question_2 = gr.Button("問題一")
                            ai_chatbot_question_3 = gr.Button("問題一")
                            create_questions_btn = gr.Button("生成問題", variant="primary")
                        ai_chatbot_audio_input = gr.Audio(sources=["microphone"], type="filepath", max_length=60, label="語音輸入")
                    with gr.Row():
                        ai_msg = gr.Textbox(label="訊息輸入",scale=3)
                        ai_send_button = gr.Button("送出", variant="primary",scale=1)
                        ai_send_feedback_btn = gr.Button("提問力回饋", variant="primary", scale=1, visible=False)          
        with gr.Tab("文章模式"):
            with gr.Row():
                reading_passage = gr.Markdown(show_label=False, latex_delimiters = [{"left": "$", "right": "$", "display": False}])
                reading_passage_speak_button = gr.Button("Speak", visible=False)
                reading_passage_audio_output = gr.Audio(label="Audio Output", visible=False) 
        with gr.Tab("重點摘要"):
            with gr.Row():
                df_summarise = gr.Markdown(show_label=False, latex_delimiters = [{"left": "$", "right": "$", "display": False}])
        with gr.Tab("心智圖",elem_id="mind_map_tab"):
            with gr.Row():
                mind_map_html = gr.HTML()
        with gr.Tab("關鍵時刻"):
            with gr.Row():
                key_moments_html = gr.HTML(value="")
        with gr.Tab("教學備課"):
            with gr.Row():
                content_subject = gr.Dropdown(label="選擇主題", choices=["數學", "自然", "國文", "英文", "社會","物理", "化學", "生物", "地理", "歷史", "公民"], value="", visible=False)             
                content_grade = gr.Dropdown(label="選擇年級", choices=["一年級", "二年級", "三年級", "四年級", "五年級", "六年級", "七年級", "八年級", "九年級", "十年級", "十一年級", "十二年級"], value="", visible=False)
                content_level = gr.Dropdown(label="差異化教學", choices=["基礎", "中級", "進階"], value="基礎")
            with gr.Row():
                with gr.Tab("學習單"):
                    with gr.Row():
                        with gr.Column(scale=1):
                            with gr.Row():
                                worksheet_content_type_name = gr.Textbox(value="worksheet", visible=False)
                                worksheet_algorithm = gr.Dropdown(label="選擇教學策略或理論", choices=["Bloom認知階層理論", "Polya數學解題法", "CRA教學法"], value="Bloom認知階層理論", visible=False)
                                worksheet_content_btn = gr.Button("生成學習單 📄", variant="primary", visible=True)
                            with gr.Accordion("微調", open=False):
                                worksheet_result_fine_tune_prompt = gr.Textbox(label="根據結果，輸入你想更改的想法")
                                worksheet_result_fine_tune_btn = gr.Button("微調結果", variant="primary")
                                worksheet_result_retrun_original = gr.Button("返回原始結果")
                            with gr.Accordion("prompt", open=False) as worksheet_accordion:
                                worksheet_prompt = gr.Textbox(label="worksheet_prompt", show_copy_button=True, lines=40)
                        with gr.Column(scale=2):
                            # 生成對應不同模式的結果
                            worksheet_result_prompt = gr.Textbox(visible=False)
                            worksheet_result_original = gr.Textbox(visible=False)
                            worksheet_result = gr.Markdown(label="初次生成結果", latex_delimiters = [{"left": "$", "right": "$", "display": False}])
                            worksheet_download_button = gr.Button("轉成 word，完成後請點擊右下角 download 按鈕", variant="primary")
                            worksheet_result_word_link = gr.File(label="Download Word")
                with gr.Tab("教案"):
                    with gr.Row():
                        with gr.Column(scale=1):
                            with gr.Row():
                                lesson_plan_content_type_name = gr.Textbox(value="lesson_plan", visible=False)                    
                                lesson_plan_time = gr.Slider(label="選擇課程時間(分鐘)", minimum=10, maximum=120, step=5, value=40)
                                lesson_plan_btn = gr.Button("生成教案 📕", variant="primary", visible=True)
                            with gr.Accordion("微調", open=False):
                                lesson_plan_result_fine_tune_prompt = gr.Textbox(label="根據結果，輸入你想更改的想法")
                                lesson_plan_result_fine_tune_btn = gr.Button("微調結果", variant="primary")
                                lesson_plan_result_retrun_original = gr.Button("返回原始結果")
                            with gr.Accordion("prompt", open=False) as lesson_plan_accordion:
                                lesson_plan_prompt = gr.Textbox(label="worksheet_prompt", show_copy_button=True, lines=40)
                        with gr.Column(scale=2):
                            # 生成對應不同模式的結果
                            lesson_plan_result_prompt = gr.Textbox(visible=False)
                            lesson_plan_result_original = gr.Textbox(visible=False)
                            lesson_plan_result = gr.Markdown(label="初次生成結果", latex_delimiters = [{"left": "$", "right": "$", "display": False}])
                            
                            lesson_plan_download_button = gr.Button("轉成 word，完成後請點擊右下角 download 按鈕", variant="primary")
                            lesson_plan_result_word_link = gr.File(label="Download Word")
                with gr.Tab("出場券"):
                    with gr.Row():
                        with gr.Column(scale=1):
                            with gr.Row():
                                exit_ticket_content_type_name = gr.Textbox(value="exit_ticket", visible=False)
                                exit_ticket_time = gr.Slider(label="選擇出場券時間(分鐘)", minimum=5, maximum=10, step=1, value=8)
                                exit_ticket_btn = gr.Button("生成出場券 🎟️", variant="primary", visible=True)
                            with gr.Accordion("微調", open=False):
                                exit_ticket_result_fine_tune_prompt = gr.Textbox(label="根據結果，輸入你想更改的想法")
                                exit_ticket_result_fine_tune_btn = gr.Button("微調結果", variant="primary")
                                exit_ticket_result_retrun_original = gr.Button("返回原始結果")
                            with gr.Accordion("prompt", open=False) as exit_ticket_accordion:
                                exit_ticket_prompt = gr.Textbox(label="worksheet_prompt", show_copy_button=True, lines=40)
                        with gr.Column(scale=2):
                            # 生成對應不同模式的結果
                            exit_ticket_result_prompt = gr.Textbox(visible=False)
                            exit_ticket_result_original = gr.Textbox(visible=False)
                            exit_ticket_result = gr.Markdown(label="初次生成結果", latex_delimiters = [{"left": "$", "right": "$", "display": False}])
                            
                            exit_ticket_download_button = gr.Button("轉成 word，完成後請點擊右下角 download 按鈕", variant="primary")
                            exit_ticket_result_word_link = gr.File(label="Download Word")
    
    
                    # with gr.Tab("素養導向閱讀題組"):
                    #     literacy_oriented_reading_content = gr.Textbox(label="輸入閱讀材料")
                    #     literacy_oriented_reading_content_btn = gr.Button("生成閱讀理解題")
    
                    # with gr.Tab("自我評估"):
                    #     self_assessment_content = gr.Textbox(label="輸入自評問卷或檢查表")
                    #     self_assessment_content_btn = gr.Button("生成自評問卷")
                    # with gr.Tab("自我反思評量"):
                    #     self_reflection_content = gr.Textbox(label="輸入自我反思活動")
                    #     self_reflection_content_btn = gr.Button("生成自我反思活動")
                    # with gr.Tab("後設認知"):
                    #     metacognition_content = gr.Textbox(label="輸入後設認知相關問題")
                    #     metacognition_content_btn = gr.Button("生成後設認知問題")
    
        with gr.Accordion("免責聲明", open=True):
            gr.Markdown("""
                本內容由AI解析並自動生成，均一平台與影片作者不對內容的準確性做出保證。建議在學習或應用前，先行查證並確認相關資訊的正確性。
                相關規範請參考：https://www.junyiacademy.org/event/jutor-policy/
            """)
    
        with gr.Accordion("See Details", open=False) as see_details:
            with gr.Row():
                is_env_prod = gr.Checkbox(value=False, label="is_env_prod")
                LLM_model = gr.Dropdown(label="LLM Model", choices=["open-ai-gpt-4o", "anthropic-claude-3-sonnet", "gemini-1.5-pro", "gemini-1.5-flash"], value="open-ai-gpt-4o", visible=True, interactive=True)
            with gr.Tab("逐字稿本文"):
                with gr.Row() as transcript_admmin:
                    transcript_kind = gr.Textbox(value="transcript", show_label=False)
                    transcript_get_button = gr.Button("取得", size="sm", variant="primary")
                    transcript_edit_button = gr.Button("編輯", size="sm", variant="primary")
                    transcript_update_button = gr.Button("儲存", size="sm", variant="primary")
                    transcript_delete_button = gr.Button("刪除", size="sm", variant="primary")
                    transcript_create_button = gr.Button("重建", size="sm", variant="primary")
                with gr.Row():
                    df_string_output = gr.Textbox(lines=40, label="Data Text", interactive=False, show_copy_button=True)
            with gr.Tab("文章本文"):
                with gr.Row() as reading_passage_admin:
                    with gr.Column():
                        with gr.Row():
                            reading_passage_kind = gr.Textbox(value="reading_passage_latex", show_label=False)
                        with gr.Row():
                            # reading_passage_text_to_latex = gr.Button("新增 LaTeX", size="sm", variant="primary")
                            reading_passage_get_button = gr.Button("取得", size="sm", variant="primary")
                            reading_passage_edit_button = gr.Button("編輯", size="sm", variant="primary")
                            reading_passage_update_button = gr.Button("儲存", size="sm", variant="primary")
                            reading_passage_delete_button = gr.Button("刪除", size="sm", variant="primary")
                            reading_passage_create_button = gr.Button("重建", size="sm", variant="primary")
                with gr.Row():
                    reading_passage_text = gr.Textbox(label="reading_passage_latex", lines=40, interactive=False, show_copy_button=True)
            with gr.Tab("重點摘要本文"):
                with gr.Row() as summary_admmin:
                    with gr.Column():
                        with gr.Row():
                            summary_kind = gr.Textbox(value="summary_markdown", show_label=False)
                        with gr.Row():
                            # summary_to_markdown = gr.Button("新增 Markdown", size="sm", variant="primary")
                            summary_get_button = gr.Button("取得", size="sm", variant="primary")
                            summary_edit_button = gr.Button("編輯", size="sm", variant="primary")
                            summary_update_button = gr.Button("儲存", size="sm", variant="primary")
                            summary_delete_button = gr.Button("刪除", size="sm", variant="primary")
                            summary_create_button = gr.Button("重建", size="sm", variant="primary")
                with gr.Row():
                    summary_text = gr.Textbox(label="summary_markdown", lines=40, interactive=False, show_copy_button=True)
            with gr.Tab("關鍵時刻本文"):
                with gr.Row() as key_moments_admin:
                    key_moments_kind = gr.Textbox(value="key_moments", show_label=False)
                    key_moments_get_button = gr.Button("取得", size="sm", variant="primary")
                    key_moments_edit_button = gr.Button("編輯", size="sm", variant="primary")
                    key_moments_update_button = gr.Button("儲存", size="sm", variant="primary")
                    key_moments_delete_button = gr.Button("刪除", size="sm", variant="primary")
                    key_moments_create_button = gr.Button("重建", size="sm", variant="primary")
                with gr.Row():
                    key_moments = gr.Textbox(label="Key Moments", lines=40, interactive=False, show_copy_button=True)
            with gr.Tab("問題本文"):
                with gr.Row() as question_list_admin:
                    questions_kind = gr.Textbox(value="questions", show_label=False)
                    questions_get_button = gr.Button("取得", size="sm", variant="primary")
                    questions_edit_button = gr.Button("編輯", size="sm", variant="primary")
                    questions_update_button = gr.Button("儲存", size="sm", variant="primary")
                    questions_delete_button = gr.Button("刪除", size="sm", variant="primary")
                    questions_create_button = gr.Button("重建", size="sm", variant="primary")
                with gr.Row():
                    questions_json = gr.Textbox(label="Questions", lines=40, interactive=False, show_copy_button=True)
            with gr.Tab("問題答案本文"):
                with gr.Row() as questions_answers_admin:
                    questions_answers_kind = gr.Textbox(value="questions_answers", show_label=False)
                    questions_answers_get_button = gr.Button("取得", size="sm", variant="primary")
                    questions_answers_edit_button = gr.Button("編輯", size="sm", variant="primary")
                    questions_answers_update_button = gr.Button("儲存", size="sm", variant="primary")
                    questions_answers_delete_button = gr.Button("刪除", size="sm", variant="primary")
                    questions_answers_create_button = gr.Button("重建", size="sm", variant="primary")
                with gr.Row():
                    questions_answers_json = gr.Textbox(label="Questions Answers", lines=40, interactive=False, show_copy_button=True)
            with gr.Tab("教學備課"):
                with gr.Row() as worksheet_admin:
                    worksheet_kind = gr.Textbox(value="ai_content_list", show_label=False)
                    worksheet_get_button = gr.Button("取得", size="sm", variant="primary")
                    worksheet_edit_button = gr.Button("編輯", size="sm", variant="primary")
                    worksheet_update_button = gr.Button("儲存", size="sm", variant="primary")
                    worksheet_delete_button = gr.Button("刪除", size="sm", variant="primary")
                    worksheet_create_button = gr.Button("重建(X)", size="sm", variant="primary", interactive=False)
                with gr.Row():
                    worksheet_json = gr.Textbox(label="worksheet", lines=40, interactive=False, show_copy_button=True)      
            with gr.Tab("逐字稿"):
                simple_html_content = gr.HTML(label="Simple Transcript")
            with gr.Tab("圖文"):
                transcript_html = gr.HTML(label="YouTube Transcript and Video") 
            with gr.Tab("markdown"):
                gr.Markdown("## 請複製以下 markdown 並貼到你的心智圖工具中，建議使用：https://markmap.js.org/repl")
                mind_map = gr.Textbox(container=True, show_copy_button=True, lines=40, elem_id="mind_map_markdown") 
            with gr.Accordion("refresh all", open=False):
                with gr.Row():
                    gr.Markdown("## 清單影片：重新生成所有內容")
                with gr.Row():
                    refresh_video_ids = gr.Textbox(label="輸入影片 id，以 , 逗號分隔")
                    refresh_btn = gr.Button("refresh", variant="primary")
                with gr.Row():
                    refresh_result = gr.JSON()
    
                    refresh_btn.click(
                        lambda: gr.update(interactive=False),
                        inputs=[],
                        outputs=[refresh_btn]
                    ).then(
                        refresh_video_LLM_all_content,
                        inputs=[refresh_video_ids],
                        outputs=[refresh_result]
                    )
                
            
        # OPEN AI CHATBOT SELECT
        chatbot_select_outputs=[
            chatbot_select_accordion, 
            all_chatbot_select_btn, 
            chatbot_open_ai_streaming, 
            chatbot_ai, 
            ai_name,
            ai_chatbot_ai_type,
            ai_chatbot_thread_id
        ]
        # 聊天机器人的配置数据
        chatbots = [
            {
                "button": vaitor_chatbot_select_btn,
                "name_state": chatbot_open_ai_name,
                "avatar_images": vaitor_chatbot_avatar_images,
                "description_value": vaitor_chatbot_description_value,
                "chatbot_select_outputs": chatbot_select_outputs,
                "chatbot_output": ai_chatbot
            },
            {
                "button": foxcat_chatbot_select_btn,
                "name_state": foxcat_chatbot_name,
                "avatar_images": foxcat_avatar_images,
                "description_value": foxcat_chatbot_description_value,
                "chatbot_select_outputs": chatbot_select_outputs,
                "chatbot_output": ai_chatbot
            },
            {
                "button": lili_chatbot_select_btn,
                "name_state": lili_chatbot_name,
                "avatar_images": lili_avatar_images,
                "description_value": lili_chatbot_description_value,
                "chatbot_select_outputs": chatbot_select_outputs,
                "chatbot_output": ai_chatbot
            },
            {
                "button": maimai_chatbot_select_btn,
                "name_state": maimai_chatbot_name,
                "avatar_images": maimai_avatar_images,
                "description_value": maimai_chatbot_description_value,
                "chatbot_select_outputs": chatbot_select_outputs,
                "chatbot_output": ai_chatbot
            }
        ]
    
        def setup_chatbot_select_button(chatbot_dict):
            button = chatbot_dict["button"]
            chatbot_name_state = chatbot_dict["name_state"]
            avatar_images = chatbot_dict["avatar_images"]
            description_value = chatbot_dict["description_value"]
            chatbot_select_outputs = chatbot_dict["chatbot_select_outputs"]
            chatbot_output = chatbot_dict["chatbot_output"]
            button.click(
                chatbot_select,  # 你可能需要修改这个函数以适应当前的逻辑
                inputs=[chatbot_name_state],
                outputs=chatbot_select_outputs
            ).then(
                update_avatar_images,
                inputs=[avatar_images, description_value],
                outputs=[chatbot_output],
                scroll_to_output=True
            )
    
        for chatbot_dict in chatbots:
            setup_chatbot_select_button(chatbot_dict)
    
        # STREAMING CHATBOT SELECT
        chatbot_open_ai_streaming_select_btn.click(
            chatbot_select,
            inputs=[chatbot_open_ai_streaming_name],
            outputs=chatbot_select_outputs
        ).then(
            create_thread_id,
            inputs=[],
            outputs=[streaming_chat_thread_id_state]
        )
    
        # ALL CHATBOT SELECT LIST
        all_chatbot_select_btn.click(
            show_all_chatbot_accordion,
            inputs=[],
            outputs=[chatbot_select_accordion, all_chatbot_select_btn]
        )
        # OPENAI ASSISTANT CHATBOT 連接按鈕點擊事件
        def setup_question_button_click(button, inputs_list, outputs_list, chat_func, scroll_to_output=True):
            button.click(
            chat_func,
            inputs=inputs_list,
            outputs=outputs_list,
            scroll_to_output=scroll_to_output
        )
    
        # 其他精靈 ai_chatbot 模式
        ai_send_button.click(
            chat_with_any_ai,
            inputs=[ai_chatbot_ai_type, password, video_id, user_data, trascript_state, key_moments, ai_msg, ai_chatbot, content_subject, content_grade, questions_answers_json, ai_chatbot_socratic_mode_btn, ai_chatbot_thread_id, ai_name],
            outputs=[ai_msg, ai_chatbot, ai_send_button, ai_send_feedback_btn, ai_chatbot_thread_id],
            scroll_to_output=True
        )
        ai_send_feedback_btn.click(
            feedback_with_ai,
            inputs=[user_data, ai_chatbot_ai_type, ai_chatbot, ai_chatbot_thread_id],
            outputs=[ai_chatbot, ai_send_feedback_btn],
            scroll_to_output=True
        )
        # 其他精靈 ai_chatbot 连接 QA 按钮点击事件
        ai_chatbot_question_buttons = [ai_chatbot_question_1, ai_chatbot_question_2, ai_chatbot_question_3]
        for question_btn in ai_chatbot_question_buttons:
            inputs_list = [ai_chatbot_ai_type, password, video_id, user_data, trascript_state, key_moments, question_btn, ai_chatbot, content_subject, content_grade, questions_answers_json, ai_chatbot_socratic_mode_btn, ai_chatbot_thread_id, ai_name]
            outputs_list = [ai_msg, ai_chatbot, ai_send_button, ai_send_feedback_btn, ai_chatbot_thread_id]
            setup_question_button_click(question_btn, inputs_list, outputs_list, chat_with_any_ai)
    
        # 為生成問題按鈕設定特殊的點擊事件
        question_buttons = [
            ai_chatbot_question_1,
            ai_chatbot_question_2,
            ai_chatbot_question_3
        ]
        create_questions_btn.click(
            change_questions, 
            inputs=[password, df_string_output], 
            outputs=question_buttons
        )
        ai_chatbot_audio_input.change(
            process_open_ai_audio_to_chatbot,
            inputs=[password, ai_chatbot_audio_input],
            outputs=[ai_msg]
        )
    
        # 当输入 YouTube 链接时触发
        process_youtube_link_inputs = [password, youtube_link, LLM_model]
        process_youtube_link_outputs = [
            video_id,
            questions_answers_json,
            df_string_output, 
            summary_text,
            df_summarise,
            key_moments,
            key_moments_html,
            mind_map, 
            mind_map_html,
            transcript_html,
            simple_html_content, 
            reading_passage_text,
            reading_passage,
            content_subject,
            content_grade,
        ]
        update_state_inputs = [
            content_subject,
            content_grade,
            df_string_output,
            key_moments,
            questions_answers_json,
        ]
        update_state_outputs = [
            content_subject_state, 
            content_grade_state, 
            trascript_state,
            key_moments_state,
            streaming_chat_thread_id_state,
            ai_chatbot_question_1,
            ai_chatbot_question_2,
            ai_chatbot_question_3
        ]
    
        youtube_link.input(
            process_youtube_link, 
            inputs=process_youtube_link_inputs, 
            outputs=process_youtube_link_outputs
        ).then(
            update_state,
            inputs=update_state_inputs,
            outputs=update_state_outputs
        )
    
        youtube_link_btn.click(
            process_youtube_link, 
            inputs=process_youtube_link_inputs, 
            outputs=process_youtube_link_outputs
        ).then(
            update_state,
            inputs=update_state_inputs,
            outputs=update_state_outputs
        )
    
    
        # --- CRUD admin ---
        def setup_content_buttons(buttons_config):
            for config in buttons_config:
                button = config['button']
                action = config['action']
                inputs = config['inputs']
                outputs = config['outputs']
                button.click(
                    fn=action,
                    inputs=inputs,
                    outputs=outputs
                )
    
        content_buttons_config = [
            # Transcript actions
            {
                'button': transcript_get_button,
                'action': get_LLM_content,
                'inputs': [video_id, transcript_kind],
                'outputs': [df_string_output]
            },
            {
                'button': transcript_create_button,
                'action': create_LLM_content,
                'inputs': [video_id, df_string_output, transcript_kind, LLM_model],
                'outputs': [df_string_output]
            },
            {
                'button': transcript_delete_button,
                'action': delete_LLM_content,
                'inputs': [video_id, transcript_kind],
                'outputs': [df_string_output]
            },
            {
                'button': transcript_edit_button,
                'action': enable_edit_mode,
                'inputs': [],
                'outputs': [df_string_output]
            },
            {
                'button': transcript_update_button,
                'action': update_LLM_content,
                'inputs': [video_id, df_string_output, transcript_kind],
                'outputs': [df_string_output]
            },
            # Reading passage actions
            {
                'button': reading_passage_get_button,
                'action': get_LLM_content,
                'inputs': [video_id, reading_passage_kind],
                'outputs': [reading_passage_text]
            },
            {
                'button': reading_passage_create_button,
                'action': create_LLM_content,
                'inputs': [video_id, df_string_output, reading_passage_kind, LLM_model],
                'outputs': [reading_passage_text]
            },
            {
                'button': reading_passage_delete_button,
                'action': delete_LLM_content,
                'inputs': [video_id, reading_passage_kind],
                'outputs': [reading_passage_text]
            },
            {
                'button': reading_passage_edit_button,
                'action': enable_edit_mode,
                'inputs': [],
                'outputs': [reading_passage_text]
            },
            {
                'button': reading_passage_update_button,
                'action': update_LLM_content,
                'inputs': [video_id, reading_passage_text, reading_passage_kind],
                'outputs': [reading_passage_text]
            },
            # Summary actions
            {
                'button': summary_get_button,
                'action': get_LLM_content,
                'inputs': [video_id, summary_kind],
                'outputs': [summary_text]
            },
            {
                'button': summary_create_button,
                'action': create_LLM_content,
                'inputs': [video_id, df_string_output, summary_kind, LLM_model],
                'outputs': [summary_text]
            },
            {
                'button': summary_delete_button,
                'action': delete_LLM_content,
                'inputs': [video_id, summary_kind],
                'outputs': [summary_text]
            },
            {
                'button': summary_edit_button,
                'action': enable_edit_mode,
                'inputs': [],
                'outputs': [summary_text]
            },
            {
                'button': summary_update_button,
                'action': update_LLM_content,
                'inputs': [video_id, summary_text, summary_kind],
                'outputs': [summary_text]
            },
            # Key moments actions
            {
                'button': key_moments_get_button,
                'action': get_LLM_content,
                'inputs': [video_id, key_moments_kind],
                'outputs': [key_moments]
            },
            {
                'button': key_moments_create_button,
                'action': create_LLM_content,
                'inputs': [video_id, df_string_output, key_moments_kind, LLM_model],
                'outputs': [key_moments]
            },
            {
                'button': key_moments_delete_button,
                'action': delete_LLM_content,
                'inputs': [video_id, key_moments_kind],
                'outputs': [key_moments]
            },
            {
                'button': key_moments_edit_button,
                'action': enable_edit_mode,
                'inputs': [],
                'outputs': [key_moments]
            },
            {
                'button': key_moments_update_button,
                'action': update_LLM_content,
                'inputs': [video_id, key_moments, key_moments_kind],
                'outputs': [key_moments]
            },
            # Questions actions
            {
                'button': questions_get_button,
                'action': get_LLM_content,
                'inputs': [video_id, questions_kind],
                'outputs': [questions_json]
            },
            {
                'button': questions_create_button,
                'action': create_LLM_content,
                'inputs': [video_id, df_string_output, questions_kind, LLM_model],
                'outputs': [questions_json]
            },
            {
                'button': questions_delete_button,
                'action': delete_LLM_content,
                'inputs': [video_id, questions_kind],
                'outputs': [questions_json]
            },
            {
                'button': questions_edit_button,
                'action': enable_edit_mode,
                'inputs': [],
                'outputs': [questions_json]
            },
            {
                'button': questions_update_button,
                'action': update_LLM_content,
                'inputs': [video_id, questions_json, questions_kind],
                'outputs': [questions_json]
            },
            # Questions answers actions
            {
                'button': questions_answers_get_button,
                'action': get_LLM_content,
                'inputs': [video_id, questions_answers_kind],
                'outputs': [questions_answers_json]
            },
            {
                'button': questions_answers_create_button,
                'action': create_LLM_content,
                'inputs': [video_id, df_string_output, questions_answers_kind, LLM_model],
                'outputs': [questions_answers_json]
            },
            {
                'button': questions_answers_delete_button,
                'action': delete_LLM_content,
                'inputs': [video_id, questions_answers_kind],
                'outputs': [questions_answers_json]
            },
            {
                'button': questions_answers_edit_button,
                'action': enable_edit_mode,
                'inputs': [],
                'outputs': [questions_answers_json]
            },
            {
                'button': questions_answers_update_button,
                'action': update_LLM_content,
                'inputs': [video_id, questions_answers_json, questions_answers_kind],
                'outputs': [questions_answers_json]
            },
            # Worksheet actions
            {
                'button': worksheet_get_button,
                'action': get_LLM_content,
                'inputs': [video_id, worksheet_kind],
                'outputs': [worksheet_json]
            },
            {
                'button': worksheet_create_button,
                'action': create_LLM_content,
                'inputs': [video_id, df_string_output, worksheet_kind, LLM_model],
                'outputs': [worksheet_json]
            },
            {
                'button': worksheet_delete_button,
                'action': delete_LLM_content,
                'inputs': [video_id, worksheet_kind],
                'outputs': [worksheet_json]
            },
            {
                'button': worksheet_edit_button,
                'action': enable_edit_mode,
                'inputs': [],
                'outputs': [worksheet_json]
            },
            {
                'button': worksheet_update_button,
                'action': update_LLM_content,
                'inputs': [video_id, worksheet_json, worksheet_kind],
                'outputs': [worksheet_json]
            },
        ]
        setup_content_buttons(content_buttons_config)
    
        # --- Education Material ---
        def setup_education_buttons(buttons_config):
            for config in buttons_config:
                button = config["button"]
                action = config["action"]
                inputs = config["inputs"]
                outputs = config["outputs"]
                button.click(
                    fn=action,
                    inputs=inputs,
                    outputs=outputs
                )
        education_buttons_config = [
            # 學習單相關按鈕
            {
                "button": worksheet_content_btn,
                "action": get_ai_content,
                "inputs": [password, user_data, video_id, df_string_output, content_subject, content_grade, content_level, worksheet_algorithm, worksheet_content_type_name],
                "outputs": [worksheet_result_original, worksheet_result, worksheet_prompt, worksheet_result_prompt]
            },
            {
                "button": worksheet_result_fine_tune_btn,
                "action": generate_ai_content_fine_tune_result,
                "inputs": [password, user_data, worksheet_result_prompt, df_string_output, worksheet_result, worksheet_result_fine_tune_prompt, worksheet_content_type_name],
                "outputs": [worksheet_result]
            },
            {
                "button": worksheet_download_button,
                "action": download_exam_result,
                "inputs": [worksheet_result],
                "outputs": [worksheet_result_word_link]
            },
            {
                "button": worksheet_result_retrun_original,
                "action": return_original_exam_result,
                "inputs": [worksheet_result_original],
                "outputs": [worksheet_result]
            },
            # 教案相關按鈕
            {
                "button": lesson_plan_btn,
                "action": get_ai_content,
                "inputs": [password, user_data, video_id, df_string_output, content_subject, content_grade, content_level, lesson_plan_time, lesson_plan_content_type_name],
                "outputs": [lesson_plan_result_original, lesson_plan_result, lesson_plan_prompt, lesson_plan_result_prompt]
            },
            {
                "button": lesson_plan_result_fine_tune_btn,
                "action": generate_ai_content_fine_tune_result,
                "inputs": [password, user_data, lesson_plan_result_prompt, df_string_output, lesson_plan_result, lesson_plan_result_fine_tune_prompt, lesson_plan_content_type_name],
                "outputs": [lesson_plan_result]
            },
            {
                "button": lesson_plan_download_button,
                "action": download_exam_result,
                "inputs": [lesson_plan_result],
                "outputs": [lesson_plan_result_word_link]
            },
            {
                "button": lesson_plan_result_retrun_original,
                "action": return_original_exam_result,
                "inputs": [lesson_plan_result_original],
                "outputs": [lesson_plan_result]
            },
            # 出場券相關按鈕
            {
                "button": exit_ticket_btn,
                "action": get_ai_content,
                "inputs": [password, user_data, video_id, df_string_output, content_subject, content_grade, content_level, exit_ticket_time, exit_ticket_content_type_name],
                "outputs": [exit_ticket_result_original, exit_ticket_result, exit_ticket_prompt, exit_ticket_result_prompt]
            },
            {
                "button": exit_ticket_result_fine_tune_btn,
                "action": generate_ai_content_fine_tune_result,
                "inputs": [password, user_data, exit_ticket_result_prompt, df_string_output, exit_ticket_result, exit_ticket_result_fine_tune_prompt, exit_ticket_content_type_name],
                "outputs": [exit_ticket_result]
            },
            {
                "button": exit_ticket_download_button,
                "action": download_exam_result,
                "inputs": [exit_ticket_result],
                "outputs": [exit_ticket_result_word_link]
            },
            {
                "button": exit_ticket_result_retrun_original,
                "action": return_original_exam_result,
                "inputs": [exit_ticket_result_original],
                "outputs": [exit_ticket_result]
            }
        ]
        setup_education_buttons(education_buttons_config)
    
        # init_params
        init_outputs = [
            admin, 
            reading_passage_admin, 
            summary_admmin, 
            see_details,
            worksheet_accordion, 
            lesson_plan_accordion, 
            exit_ticket_accordion,
            password, 
            youtube_link,
            block_ready_flag,
            chatbot_open_ai_streaming,
            chatbot_ai,
            ai_chatbot_params,
            is_env_prod,
        ]
        demo.load(
            init_params, 
            inputs =[youtube_link],
            outputs = init_outputs
        )

    app = gr.mount_gradio_app(
        app, demo, "/", server_name="0.0.0.0", server_port=7860, show_error=True
    )
    return app

if __name__ == "__main__":
    import uvicorn

    app = create_app()
    uvicorn.run(app, host="0.0.0.0", port=7860)