app.py

import os
import gradio as gr
import asyncio
from langchain_core.prompts import PromptTemplate
from langchain_community.document_loaders import PyPDFLoader
from langchain_google_genai import ChatGoogleGenerativeAI
import google.generativeai as genai
from langchain.chains.question_answering import load_qa_chain

# Initialize a dictionary to store chat history and context per session
session_contexts = {}

async def initialize(file_path, question, session_id):
    genai.configure(api_key=os.getenv("GOOGLE_API_KEY"))
    model = genai.GenerativeModel('gemini-pro')
    model = ChatGoogleGenerativeAI(model="gemini-pro", temperature=0.3)
    
    # Refined prompt template to encourage precise and concise answers
    prompt_template = """You are a helpful assistant. Use the context provided below to answer the question precisely and concisely. 
                          If the answer is not contained in the context, respond with "answer not available in context".

                          Context:
                          {context}
                          
                          Conversation History:
                          {history}
                          
                          Question:
                          {question}
                          
                          Answer:
                        """
    prompt = PromptTemplate(template=prompt_template, input_variables=["context", "history", "question"])

    # Get or initialize the context and history for the current session
    context_history = session_contexts.get(session_id, {"context": "", "history": ""})
    combined_context = context_history["context"]
    conversation_history = context_history["history"]

    if os.path.exists(file_path):
        pdf_loader = PyPDFLoader(file_path)
        pages = pdf_loader.load_and_split()

        # Extract content from each page and store along with page number
        page_contexts = [page.page_content for i, page in enumerate(pages)]
        context = "\n".join(page_contexts[:30])  # Using the first 30 pages for context

        # Load the question-answering chain
        stuff_chain = load_qa_chain(model, chain_type="stuff", prompt=prompt)

        # Combine previous context and conversation history with the new context
        full_context = combined_context + "\n" + context
        full_history = conversation_history + f"\nQ: {question}\nA: {answer}"
        
        # Get the answer from the model
        stuff_answer = await stuff_chain.ainvoke({"input_documents": pages, "question": question, "context": full_context, "history": full_history})
        answer = stuff_answer.get('output_text', '').strip()

        # Identify key sentences or phrases
        key_phrases = answer.split(". ")  # Split answer into sentences for more precise matching

        # Score each page based on the presence of key phrases
        page_scores = [0] * len(pages)
        for i, page in enumerate(pages):
            for phrase in key_phrases:
                if phrase.lower() in page.page_content.lower():
                    page_scores[i] += 1

        # Determine the top pages based on highest scores
        top_pages_with_scores = sorted(enumerate(page_scores), key=lambda x: x[1], reverse=True)
        top_pages = [i + 1 for i, score in top_pages_with_scores if score > 0][:2]  # Get top 2 pages

        # Generate links for each top page
        file_name = os.path.basename(file_path)
        page_links = [f"[Page {p}](file://{os.path.abspath(file_path)})" for p in top_pages]
        page_links_str = ', '.join(page_links)

        if top_pages:
            source_str = f"Top relevant page(s): {page_links_str}"
        else:
            source_str = "Top relevant page(s): Not found in specific page"

        # Create a clickable link for the document
        source_link = f"[Document: {file_name}](file://{os.path.abspath(file_path)})"

        # Update session context with the new question and answer
        session_contexts[session_id] = {
            "context": full_context,
            "history": full_history + f"\nQ: {question}\nA: {answer}"
        }

        return f"Answer: {answer}\n{source_str}\n{source_link}"
    else:
        return "Error: Unable to process the document. Please ensure the PDF file is valid."

# Define Gradio Interface for QA and Chat History
input_file = gr.File(label="Upload PDF File")
input_question = gr.Textbox(label="Ask about the document")
output_text = gr.Textbox(label="Answer and Top Pages", lines=10, max_lines=10)

def get_chat_history(session_id):
    if session_id in session_contexts:
        return session_contexts[session_id]["history"]
    else:
        return "No history available for this session."

async def pdf_qa(file, question, session_id):
    if file is None:
        return "Error: No file uploaded. Please upload a PDF document."
    
    answer = await initialize(file.name, question, session_id)
    return answer

# Create Gradio Interfaces
qa_interface = gr.Interface(
    fn=lambda file, question, session_id: asyncio.run(pdf_qa(file, question, session_id)),
    inputs=[input_file, input_question, gr.Textbox(label="Session ID", placeholder="Enter a session ID to track your conversation")],
    outputs=output_text,
    title="PDF Question Answering System",
    description="Upload a PDF file and ask questions about the content. Provide a session ID to maintain conversation context."
)

history_interface = gr.Interface(
    fn=lambda session_id: get_chat_history(session_id),
    inputs=gr.Textbox(label="Session ID", placeholder="Enter a session ID to view chat history"),
    outputs=gr.Textbox(label="Chat History", lines=20, max_lines=20),
    title="Chat History",
    description="View the history of interactions for a specific session."
)

# Launch both interfaces
qa_interface.launch(share=True)
history_interface.launch(share=True)