app.py

import os
import requests
import streamlit as st
from io import BytesIO
from PyPDF2 import PdfReader
from langchain.text_splitter import RecursiveCharacterTextSplitter
from langchain.embeddings import HuggingFaceEmbeddings
from langchain.vectorstores import FAISS
from transformers import pipeline
import torch

# Update imports to reflect the new locations
from langchain_community.embeddings import HuggingFaceEmbeddings
from langchain_community.vectorstores import FAISS

# Set up the page configuration as the first Streamlit command
st.set_page_config(page_title="RAG-based PDF Chat", layout="centered", page_icon="📄")

# Initialize embedding function with the new import
embedding_function = HuggingFaceEmbeddings(model_name="sentence-transformers/all-MiniLM-L6-v2")

# Load the question-answering pipeline model
@st.cache_resource
def load_qa_pipeline():
    qa_pipeline = pipeline("question-answering", model="distilbert-base-cased-distilled-squad")  # Use a text-based QA model
    return qa_pipeline

qa_pipeline = load_qa_pipeline()

# Generate summary based on the retrieved text
def generate_summary_with_qa_pipeline(query, retrieved_text):
    # Use the pipeline to directly answer questions based on document context
    summary = qa_pipeline(question=query, context=retrieved_text)
    return summary["answer"]

# List of Hugging Face PDF URLs
PDF_URLS = [
    "https://huggingface.co/spaces/tahirsher/GenAI_Lawyers_Guide/blob/main/administrator92ada0936848e501425591b4ad0cd417.pdf",
    "https://huggingface.co/spaces/tahirsher/GenAI_Lawyers_Guide/blob/main/Pakistan%20Penal%20Code.pdf",
]

# Helper function to convert Hugging Face blob URLs to direct download URLs
def get_huggingface_raw_url(url):
    if "huggingface.co" in url and "/blob/" in url:
        return url.replace("/blob/", "/resolve/")
    return url

# Fetch and extract text from PDF files hosted on Hugging Face
def fetch_pdf_text_from_huggingface(urls):
    text = ""
    for url in urls:
        raw_url = get_huggingface_raw_url(url)
        response = requests.get(raw_url)
        if response.status_code == 200:
            pdf_file = BytesIO(response.content)
            try:
                pdf_reader = PdfReader(pdf_file)
                for page in pdf_reader.pages:
                    page_text = page.extract_text()
                    if page_text:
                        text += page_text
            except Exception as e:
                st.error(f"Failed to read PDF from URL {url}: {e}")
        else:
            st.error(f"Failed to fetch PDF from URL: {url}")
    return text

# Split text into manageable chunks
@st.cache_data
def get_text_chunks(text):
    text_splitter = RecursiveCharacterTextSplitter(chunk_size=10000, chunk_overlap=1000)
    chunks = text_splitter.split_text(text)
    return chunks

# Initialize embedding function
embedding_function = HuggingFaceEmbeddings(model_name="sentence-transformers/all-MiniLM-L6-v2")

# Create a FAISS vector store with embeddings
@st.cache_resource
def load_or_create_vector_store(text_chunks):
    vector_store = FAISS.from_texts(text_chunks, embedding=embedding_function)
    return vector_store

# Generate summary based on the retrieved text
def generate_summary_with_qa_pipeline(query, retrieved_text):
    # Use the pipeline to directly answer questions based on document context
    summary = qa_pipeline(question=query, context=retrieved_text)
    return summary["answer"]  # Directly access the answer

# Generate response for user query
def user_input(user_question, vector_store):
    docs = vector_store.similarity_search(user_question)
    context_text = " ".join([doc.page_content for doc in docs])
    return generate_summary_with_qa_pipeline(user_question, context_text)

# Main function to run the Streamlit app
def main():
    st.title("📄 Gen AI Lawyers Guide")

    # Load documents from Hugging Face
    raw_text = fetch_pdf_text_from_huggingface(PDF_URLS)
    text_chunks = get_text_chunks(raw_text)
    vector_store = load_or_create_vector_store(text_chunks)

    # User question input
    user_question = st.text_input("Ask a Question:", placeholder="Type your question here...")

    if st.button("Get Response"):
        if not user_question:
            st.warning("Please enter a question before submitting.")
        else:
            with st.spinner("Generating response..."):
                answer = user_input(user_question, vector_store)
                st.markdown(f"**🤖 AI:** {answer}")

if __name__ == "__main__":
    main()