app.py

import streamlit as st
from langchain_community.vectorstores import FAISS
from langchain_community.document_loaders import TextLoader
from langchain_community.embeddings import HuggingFaceEmbeddings
from langchain.chains import RetrievalQA
from langchain_google_genai import GoogleGenerativeAI
from langchain.prompts import PromptTemplate
#from langchain.chains import load_qa_chain, RetrievalQA
import requests
from bs4 import BeautifulSoup
from urllib.parse import urljoin
import re
from collections import deque
import time
import numpy as np


# Crawling function
def crawl(start_url: str, max_depth: int = 1, delay: float = 0.1) :
    visited = set()
    results = []
    queue = deque([(start_url, 0)])
    crawled_urls = []

    while queue:
        url, depth = queue.popleft()

        if depth > max_depth or url in visited:
            continue

        visited.add(url)
        crawled_urls.append(url)

        try:
            time.sleep(delay)
            response = requests.get(url)
            soup = BeautifulSoup(response.text, 'html.parser')

            text = soup.get_text()
            text = re.sub(r'\s+', ' ', text).strip()

            results.append((url, text))

            if depth < max_depth:
                for link in soup.find_all('a', href=True):
                    next_url = urljoin(url, link['href'])
                    if next_url.startswith('https://docs.nvidia.com/cuda/') and next_url not in visited:
                        queue.append((next_url, depth + 1))
                    if len(queue) > 10:
                        break
        except Exception as e:
            print(f"Error crawling {url}: {e}")

    return results, crawled_urls


# Text chunking function
def chunk_text(text: str, max_chunk_size: int = 1000) :
    chunks = []
    current_chunk = ""

    for sentence in re.split(r'(?<=[.!?])\s+', text):
        if len(current_chunk) + len(sentence) <= max_chunk_size:
            current_chunk += sentence + " "
        else:
            chunks.append(current_chunk.strip())
            current_chunk = sentence + " "

    if current_chunk:
        chunks.append(current_chunk.strip())

    return chunks

# Streamlit UI
st.title("CUDA Documentation QA System")

# Initialize global variables
if 'vector_store' not in st.session_state:
    st.session_state.vector_store = None
if 'documents_loaded' not in st.session_state:
    st.session_state.documents_loaded = False

# Crawling and processing the data
if st.button('Crawl CUDA Documentation'):
    with st.spinner('Crawling CUDA documentation...'):
        crawled_data, crawled_urls = crawl("https://docs.nvidia.com/cuda/", max_depth=1, delay=0.1)
        st.write(f"Processed {len(crawled_data)} pages.")
        
        texts = []
        for url, text in crawled_data:
            chunks = chunk_text(text, max_chunk_size=1024)
            texts.extend(chunks)
        st.success("Crawling and processing completed.")
        
        # Create embeddings
        embeddings = HuggingFaceEmbeddings(model_name='sentence-transformers/all-MiniLM-L6-v2', model_kwargs={'device': 'cpu'})
        
        # Store embeddings in FAISS
        st.session_state.vector_store = FAISS.from_texts(texts, embeddings)
        st.session_state.documents_loaded = True
        st.write("Embeddings stored in FAISS.")

# Asking questions
query = st.text_input("Enter your question about CUDA:")
if query and st.session_state.documents_loaded:
    with st.spinner('Searching for an answer...'):
        # Initialize Google Generative AI
        llm = GoogleGenerativeAI(model='gemini-1.0-pro', google_api_key="AIzaSyC1AvHnvobbycU8XSCXh-gRq3DUfG0EP98")


        #Create a PromptTemplate for the QA chain
        qa_prompt = PromptTemplate(template="Answer the following question based on the context provided:\n\nContext: {context}\n\nQuestion: {question}\n\nAnswer:", input_variables=["context", "question"])

        # Create the retrieval QA chain
        qa_chain = RetrievalQA.from_chain_type(
            retriever=st.session_state.vector_store.as_retriever(),
            #chain_type="stuff",
            llm=llm,
            #chain_type_kwargs={"prompt": qa_prompt}
        )

        response = qa_chain({"question": query})
        st.write("**Answer:**")
        st.write(response['answer'])
        st.write("**Source:**")
        st.write(response['source'])
elif query:
    st.warning("Please crawl the CUDA documentation first.")