import gradio as gr
from inference import get_evo_response, get_gpt_response
from logger import log_feedback
import retrain
import pandas as pd
import os

def advisor_interface(query, context, file, feedback_choice):
    evo_output, evo_reasoning = get_evo_response(query, context, file)
    gpt_output = get_gpt_response(query, context)

    if feedback_choice != "No feedback":
        label = 1 if feedback_choice == "👍 Helpful" else 0
        log_feedback(query, context, evo_output, label)

    return evo_reasoning, gpt_output, load_history()

def retrain_evo():
    retrain.fine_tune_on_feedback()
    return "✅ Evo retrained on feedback.", load_history()

def load_history():
    if os.path.exists("feedback_log.csv"):
        df = pd.read_csv("feedback_log.csv")
        return df.tail(10).to_markdown(index=False)
    return "No history available yet."

with gr.Blocks() as demo:
    gr.Markdown("## 🧠 EvoRAG – Retrieval-Augmented Adaptive AI (General Purpose)")

    with gr.Row():
        query = gr.Textbox(label="📝 Ask any question", placeholder="e.g. What are the effects of inflation?")
        context = gr.Textbox(label="📂 Optional background notes or memos", placeholder="Paste policy, notes, or ideas...")

    with gr.Row():
        file = gr.File(label="📎 Upload optional file (.txt or .pdf)", file_types=[".txt", ".pdf"])

    with gr.Row():
        feedback = gr.Radio(["👍 Helpful", "👎 Not Helpful", "No feedback"], label="Was Evo’s answer useful?", value="No feedback")

    with gr.Row():
        evo_out = gr.Textbox(label="🔬 EvoRAG Suggestion (with reasoning)", lines=6)
        gpt_out = gr.Textbox(label="🤖 GPT-3.5 Suggestion", lines=6)

    run_button = gr.Button("Run Advisors")
    history_output = gr.Textbox(label="📜 Recent History")

    run_button.click(fn=advisor_interface, inputs=[query, context, file, feedback], outputs=[evo_out, gpt_out, history_output])

    gr.Markdown("---")
    gr.Markdown("### 🔁 Retrain Evo on Feedback")
    retrain_button = gr.Button("📚 Retrain Evo")
    retrain_status = gr.Textbox(label="🛠️ Retrain Status")

    retrain_button.click(fn=retrain_evo, inputs=[], outputs=[retrain_status, history_output])

demo.launch()