app.py

import gradio as gr
import PyPDF2
import io
from together import Together
from langchain_community.vectorstores import FAISS
from langchain_huggingface import HuggingFaceEmbeddings
from langchain.docstore.document import Document
from langchain.text_splitter import RecursiveCharacterTextSplitter
from langchain.chains import ConversationalRetrievalChain
from langchain.memory import ConversationBufferMemory
from langchain.llms.base import LLM
from typing import List, Optional
import traceback


# ---------------------------
# WRAP TOGETHER API AS LLM
# ---------------------------
class TogetherLLM(LLM):
    client: Together = None
    model: str = "meta-llama/Llama-3.3-70B-Instruct-Turbo"
    temperature: float = 0.3
    max_tokens: int = 1000

    def __init__(self, client, model="meta-llama/Llama-3.3-70B-Instruct-Turbo", temperature=0.3, max_tokens=1000, **kwargs):
        super().__init__(**kwargs)
        object.__setattr__(self, 'client', client)
        object.__setattr__(self, 'model', model)
        object.__setattr__(self, 'temperature', temperature)
        object.__setattr__(self, 'max_tokens', max_tokens)

    @property
    def _llm_type(self) -> str:
        return "together-llm"

    def _call(self, prompt: str, stop: Optional[List[str]] = None) -> str:
        try:
            response = self.client.chat.completions.create(
                model=self.model,
                messages=[{"role": "user", "content": prompt}],
                max_tokens=self.max_tokens,
                temperature=self.temperature,
            )
            return response.choices[0].message.content.strip()
        except Exception as e:
            return f"Error generating response: {str(e)}"

    class Config:
        arbitrary_types_allowed = True


# ---------------------------
# PDF TEXT EXTRACTION
# ---------------------------
def extract_text_from_pdf(pdf_file):
    """Extract text from PDF with page references"""
    docs = []
    try:
        print("Starting PDF extraction...")
        
        # Handle different input types
        if hasattr(pdf_file, 'name'):
            # File uploaded through Gradio
            with open(pdf_file.name, 'rb') as file:
                pdf_content = file.read()
        elif hasattr(pdf_file, "read"):
            pdf_content = pdf_file.read()
            if hasattr(pdf_file, "seek"):
                pdf_file.seek(0)
        else:
            pdf_content = pdf_file

        pdf_reader = PyPDF2.PdfReader(io.BytesIO(pdf_content))
        print(f"PDF has {len(pdf_reader.pages)} pages")

        for page_num, page in enumerate(pdf_reader.pages, start=1):
            try:
                page_text = page.extract_text()
                if page_text and page_text.strip():
                    docs.append(Document(
                        page_content=page_text.strip(), 
                        metadata={"page": page_num, "source": "financial_policy"}
                    ))
                    print(f"Extracted text from page {page_num}: {len(page_text)} characters")
                else:
                    docs.append(Document(
                        page_content="[No extractable text found on this page]",
                        metadata={"page": page_num, "source": "financial_policy"}
                    ))
            except Exception as e:
                print(f"Error extracting page {page_num}: {str(e)}")
                docs.append(Document(
                    page_content=f"[Error extracting page {page_num}: {str(e)}]",
                    metadata={"page": page_num, "source": "financial_policy"}
                ))
        
        print(f"Total documents extracted: {len(docs)}")
        return docs
        
    except Exception as e:
        print(f"Error in PDF extraction: {str(e)}")
        traceback.print_exc()
        return [Document(page_content=f"Error extracting text: {str(e)}", metadata={"page": -1})]


# ---------------------------
# BUILD KNOWLEDGE BASE (FAISS)
# ---------------------------
def build_vector_db(docs):
    """Convert extracted documents into FAISS vector DB"""
    try:
        print("Building vector database...")
        text_splitter = RecursiveCharacterTextSplitter(
            chunk_size=1000, 
            chunk_overlap=100,
            separators=["\n\n", "\n", ". ", " ", ""]
        )
        split_docs = text_splitter.split_documents(docs)
        print(f"Split into {len(split_docs)} chunks")

        # Initialize embeddings
        embeddings = HuggingFaceEmbeddings(
            model_name="sentence-transformers/all-MiniLM-L6-v2",
            model_kwargs={'device': 'cpu'}
        )
        print("Embeddings model loaded")

        # Create FAISS database
        db = FAISS.from_documents(split_docs, embeddings)
        print("Vector database created successfully")
        return db
        
    except Exception as e:
        print(f"Error building vector database: {str(e)}")
        traceback.print_exc()
        return None


# ---------------------------
# CHATBOT PIPELINE
# ---------------------------
def create_chatbot(api_key, db):
    """Set up ConversationalRetrievalChain with memory"""
    try:
        print("Creating chatbot...")
        client = Together(api_key=api_key)
        llm = TogetherLLM(client=client)

        retriever = db.as_retriever(
            search_type="similarity",
            search_kwargs={"k": 4}
        )

        memory = ConversationBufferMemory(
            memory_key="chat_history", 
            return_messages=True,
            output_key="answer"
        )

        qa_chain = ConversationalRetrievalChain.from_llm(
            llm=llm,
            retriever=retriever,
            memory=memory,
            return_source_documents=True,
            verbose=True,
        )
        print("Chatbot created successfully")
        return qa_chain
        
    except Exception as e:
        print(f"Error creating chatbot: {str(e)}")
        traceback.print_exc()
        return None


# ---------------------------
# GRADIO APP
# ---------------------------
def create_app():
    with gr.Blocks(title="📊 Financial Policy Document Chatbot", theme=gr.themes.Soft()) as app:
        gr.Markdown("# 📊 Financial Policy Document Chatbot")
        gr.Markdown("""
        Upload a financial policy PDF document and ask questions about its content. 
        The chatbot will provide answers with page references from the document.
        """)

        with gr.Row():
            with gr.Column(scale=1):
                api_key_input = gr.Textbox(
                    label="Together API Key",
                    placeholder="Enter your Together API key here...",
                    type="password",
                )
                pdf_file = gr.File(
                    label="Upload Financial Policy PDF",
                    file_types=[".pdf"],
                )
                process_button = gr.Button("📄 Process PDF", variant="primary")
                status_message = gr.Textbox(label="Status", interactive=False, lines=3)

            with gr.Column(scale=2):
                chatbot = gr.Chatbot(label="Chat with Financial Policy Document", height=500)
                with gr.Row():
                    question = gr.Textbox(
                        label="Ask a question about the document",
                        placeholder="Example: What is the budget allocation for infrastructure?",
                        lines=2,
                        scale=4
                    )
                    submit_button = gr.Button("🔍 Ask", variant="secondary", scale=1)
                
                gr.Markdown("""
                **Sample Questions:**
                - What is the debt policy outlined in the document?
                - How much budget is allocated for infrastructure?
                - What are the revenue sources mentioned?
                - What are the key financial objectives?
                """)

        # State variables
        db_state = gr.State()
        qa_chain_state = gr.State()

        # Event handlers
        def process_pdf_handler(pdf_file, api_key):
            try:
                if pdf_file is None:
                    return "⚠️ Please upload a PDF file.", None, None
                
                if not api_key or api_key.strip() == "":
                    return "⚠️ Please enter your Together API key.", None, None

                status_msg = "🔄 Processing PDF... This may take a few moments."
                yield status_msg, None, None

                # Extract text from PDF
                docs = extract_text_from_pdf(pdf_file)
                if not docs or len(docs) == 0:
                    yield "⚠️ No text could be extracted from the PDF.", None, None
                    return

                # Check if extraction was successful
                valid_docs = [doc for doc in docs if not doc.page_content.startswith("[Error") and not doc.page_content.startswith("[No extractable")]
                if len(valid_docs) == 0:
                    yield "⚠️ No readable text found in the PDF.", None, None
                    return

                status_msg = f"📄 Extracted text from {len(docs)} pages. Building search database..."
                yield status_msg, None, None

                # Build vector database
                db = build_vector_db(docs)
                if db is None:
                    yield "⚠️ Failed to build search database.", None, None
                    return

                status_msg = f"🔍 Search database created. Setting up chatbot..."
                yield status_msg, None, None

                # Create chatbot
                qa_chain = create_chatbot(api_key, db)
                if qa_chain is None:
                    yield "⚠️ Failed to create chatbot.", None, None
                    return

                final_status = f"✅ Successfully processed PDF with {len(docs)} pages. Ready to answer questions!"
                yield final_status, db, qa_chain

            except Exception as e:
                error_msg = f"❌ Error processing PDF: {str(e)}"
                print(f"Process PDF Error: {str(e)}")
                traceback.print_exc()
                yield error_msg, None, None

        def chat_handler(user_question, qa_chain, history):
            if not user_question or user_question.strip() == "":
                return history, history, ""

            if qa_chain is None:
                return history + [(user_question, "⚠️ Please process a PDF document first.")], history, ""

            try:
                # Get response from the chain
                result = qa_chain({"question": user_question})
                answer = result["answer"]

                # Add source references
                if "source_documents" in result and result["source_documents"]:
                    pages = []
                    for doc in result["source_documents"]:
                        if "page" in doc.metadata:
                            pages.append(doc.metadata["page"])
                    
                    if pages:
                        unique_pages = sorted(set(pages))
                        if len(unique_pages) == 1:
                            answer += f"\n\n📌 **Reference:** Page {unique_pages[0]}"
                        else:
                            answer += f"\n\n📌 **References:** Pages {', '.join(map(str, unique_pages))}"

                new_history = history + [(user_question, answer)]
                return new_history, new_history, ""

            except Exception as e:
                error_response = f"❌ Error processing question: {str(e)}"
                print(f"Chat Error: {str(e)}")
                traceback.print_exc()
                return history + [(user_question, error_response)], history, ""

        def clear_input():
            return ""

        # Bind events
        process_button.click(
            fn=process_pdf_handler,
            inputs=[pdf_file, api_key_input],
            outputs=[status_message, db_state, qa_chain_state],
        )

        submit_button.click(
            fn=chat_handler,
            inputs=[question, qa_chain_state, chatbot],
            outputs=[chatbot, chatbot, question],
        )

        question.submit(
            fn=chat_handler,
            inputs=[question, qa_chain_state, chatbot],
            outputs=[chatbot, chatbot, question],
        )

    return app


# ---------------------------
# MAIN EXECUTION
# ---------------------------
if __name__ == "__main__":
    app = create_app()
    app.launch(
        share=True,
        server_name="0.0.0.0",
        server_port=7860,
        debug=True
    )