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

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

# Load the summarization pipeline model
@st.cache_resource
def load_summarization_pipeline():
    summarizer = pipeline("summarization", model="facebook/bart-large-cnn")
    return summarizer

summarizer = load_summarization_pipeline()

# Dictionary of Hugging Face PDF URLs grouped by folders
PDF_FOLDERS = {
    "PPC and Administration": [
        "https://huggingface.co/spaces/tahirsher/GenAI_Lawyers_Guide/tree/main/PPC%20and%20Administration",
    ],
    "IHC": [
        "https://huggingface.co/spaces/tahirsher/GenAI_Lawyers_Guide/tree/main/IHC"
    "LHC": [
        "https://huggingface.co/spaces/tahirsher/GenAI_Lawyers_Guide/tree/main/LHC"
    "Lahore High Court Rules and Orders": [
        "https://huggingface.co/spaces/tahirsher/GenAI_Lawyers_Guide/tree/main/Lahore%20High%20Court%20Rules%20and%20Orders"
    "PHC": [
        "https://huggingface.co/spaces/tahirsher/GenAI_Lawyers_Guide/tree/main/PHC"
    "SC": [
        "https://huggingface.co/spaces/tahirsher/GenAI_Lawyers_Guide/tree/main/SC"
    ],
}

# 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 all PDFs in specified folders
def fetch_pdf_text_from_folders(pdf_folders):
    all_text = ""
    for folder_name, urls in pdf_folders.items():
        folder_text = f"\n[Folder: {folder_name}]\n"
        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:
                            folder_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}")
        all_text += folder_text
    return all_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_huggingface(query, retrieved_text):
    summarization_input = f"{query}\n\nRelated information:\n{retrieved_text}"
    max_input_length = 1024
    summarization_input = summarization_input[:max_input_length]
    summary = summarizer(summarization_input, max_length=500, min_length=50, do_sample=False)
    return summary[0]["summary_text"]

# 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_huggingface(user_question, context_text)

# Main function to run the Streamlit app
def main():
    st.title("📄 Gen AI Lawyers Guide")
    raw_text = fetch_pdf_text_from_folders(PDF_FOLDERS)
    text_chunks = get_text_chunks(raw_text)
    vector_store = load_or_create_vector_store(text_chunks)

    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()