app.py

import gradio as gr
import os
from dotenv import load_dotenv
from langchain.text_splitter import RecursiveCharacterTextSplitter
from langchain_mistralai import MistralAIEmbeddings
from langchain_groq import ChatGroq
from langchain_chroma import Chroma
from langchain.chains import RetrievalQA
from pypdf import PdfReader
from docx import Document
import time
import fitz  # PyMuPDF

# Configuration for batch processing and rate limiting
BATCH_SIZE = 20  # Reduced for embeddings (was 50)
BATCH_DELAY = 1.0  # Increased delay for API rate limiting (was 0.2)
MAX_RETRIES = 3  # Increased retries for API stability
CACHE_ENABLED = True  # Enable caching for embeddings

# Load environment variables from .env file
load_dotenv()

def process_chunks_in_batches_with_progress(chunks, metadatas, file_name, progress_callback=None):
    """Process document chunks in optimized batches with minimal delays."""
    total_chunks = len(chunks)
    processed = 0
    
    print(f"Processing {total_chunks} chunks in batches of {BATCH_SIZE}")
    if progress_callback:
        progress_callback(f"Starting optimized batch processing: {total_chunks} chunks total")
    
    for i in range(0, total_chunks, BATCH_SIZE):
        batch_end = min(i + BATCH_SIZE, total_chunks)
        batch_chunks = chunks[i:batch_end]
        batch_metadatas = metadatas[i:batch_end]
        
        # Enhanced retry logic with exponential backoff
        success = False
        for attempt in range(MAX_RETRIES):
            try:
                batch_start_time = time.time()
                vectorstore.add_texts(batch_chunks, metadatas=batch_metadatas)
                batch_time = time.time() - batch_start_time
                
                processed += len(batch_chunks)
                progress_msg = f"✓ Batch {i//BATCH_SIZE + 1}/{(total_chunks-1)//BATCH_SIZE + 1}: {len(batch_chunks)} chunks ({processed}/{total_chunks}) - {batch_time:.2f}s"
                print(progress_msg)
                if progress_callback:
                    progress_callback(progress_msg)
                success = True
                break
                
            except Exception as e:
                wait_time = min(2 ** attempt, 10)  # Exponential backoff, max 10s
                if "rate limit" in str(e).lower() or "quota" in str(e).lower():
                    print(f"Rate limit hit, waiting {wait_time}s...")
                    time.sleep(wait_time)
                elif attempt < MAX_RETRIES - 1:
                    print(f"Attempt {attempt + 1} failed, retrying in {wait_time}s...")
                    time.sleep(wait_time)
                else:
                    print(f"✗ Failed batch {i//BATCH_SIZE + 1} after {MAX_RETRIES} attempts: {str(e)[:100]}...")
        
        # Respectful delay between successful batches to avoid rate limits
        if success and batch_end < total_chunks:
            time.sleep(BATCH_DELAY)
    
    print(f"✓ Processing complete: {processed}/{total_chunks} chunks processed")
    return processed

# Initialize optimized embeddings with performance settings
try:
    embeddings = MistralAIEmbeddings(
        model="mistral-embed",  # Specify model for consistency
        # Add timeout and retry settings
    )
    print("✓ MistralAI embeddings initialized successfully")
except Exception as e:
    print(f"✗ Error initializing MistralAI embeddings: {e}")
    print("Retrying with default settings...")
    embeddings = MistralAIEmbeddings()

# Initialize optimized vector store
vectorstore = Chroma(
    collection_name="research_papers",
    embedding_function=embeddings,
    persist_directory="./chroma_db",
    # Collection metadata for optimization
    collection_metadata={"hnsw:space": "cosine"}  # Faster similarity search
)

# Initialize optimized text splitter for faster processing
text_splitter = RecursiveCharacterTextSplitter(
    chunk_size=400,  # Reduced for faster embedding (was 600)
    chunk_overlap=50,  # Minimal overlap for faster processing
    length_function=len,
    separators=["\n\n", "\n", ".", " ", ""]  # Simplified separators for speed
)

def parse_pdf(file_path):
    """Parse PDF file and extract text using PyMuPDF."""
    text = ""
    try:
        # Try PyMuPDF first (better extraction)
        doc = fitz.open(file_path)
        for page in doc:
            text += page.get_text() + "\n"
        doc.close()
    except Exception as e:
        print(f"PyMuPDF failed: {e}, falling back to pypdf")
        # Fallback to pypdf
        reader = PdfReader(file_path)
        for page in reader.pages:
            text += page.extract_text() + "\n"
    return text

def parse_docx(file_path):
    """Parse DOCX file and extract text."""
    doc = Document(file_path)
    text = ""
    for paragraph in doc.paragraphs:
        text += paragraph.text + "\n"
    return text

def process_document(file):
    """Process uploaded document and add to vector store."""
    if file is None:
        return "No file uploaded."

    # Get the file path from Gradio's file input
    file_path = file.name if hasattr(file, 'name') else str(file)

    try:
        # Determine file type and parse
        if file_path.lower().endswith('.pdf'):
            text = parse_pdf(file_path)
        elif file_path.lower().endswith('.docx'):
            text = parse_docx(file_path)
        else:
            return "Unsupported file format. Please upload PDF or DOCX files."

        # Debug: Print text length and preview
        print(f"Extracted text length: {len(text)} characters")
        print(f"Text preview: {text[:500]}...")

        # Split text into chunks
        chunks = text_splitter.split_text(text)

        # Debug: Print chunk information
        print(f"Number of chunks: {len(chunks)}")
        print(f"Average chunk length: {sum(len(chunk) for chunk in chunks) / len(chunks) if chunks else 0}")

        # Prepare metadata for chunks
        metadatas = [{"source": os.path.basename(file_path)} for _ in chunks]

        # Add to vector store in batches
        try:
            processed_chunks = process_chunks_in_batches_with_progress(chunks, metadatas, os.path.basename(file_path))
            print(f"Successfully processed {processed_chunks}/{len(chunks)} chunks")
        except Exception as e:
            print(f"Error in batch processing: {e}")
            return f"Error storing document chunks: {str(e)}"

        return f"Successfully processed {os.path.basename(file_path)} with {len(chunks)} chunks ({processed_chunks} stored). Total text length: {len(text)} characters."

    except Exception as e:
        return f"Error processing document: {str(e)}"

def query_documents(question):
    """Query the documents using RAG with performance monitoring."""
    start_time = time.time()
    
    try:
        # Check if there are documents
        if vectorstore._collection.count() == 0:
            return "No documents uploaded yet. Please upload some research papers first."
    except:
        return "No documents uploaded yet. Please upload some research papers first."

    # Initialize LLM
    try:
        llm = ChatGroq(model="llama-3.1-8b-instant")
        print("✓ Llama 3.1 8B Instant LLM initialized successfully")
    except Exception as e:
        print(f"✗ Error initializing LLM: {e}")
        return f"Error initializing language model: {str(e)}"

    # Create optimized RAG chain with faster retrieval
    retriever = vectorstore.as_retriever(
        search_type="similarity",
        search_kwargs={
            "k": 3  # Reduced back to 3 for faster retrieval
        }
    )
    
    qa_chain = RetrievalQA.from_chain_type(
        llm=llm,
        chain_type="stuff",
        retriever=retriever,
        return_source_documents=False  # Faster response
    )

    # Get answer with error handling and timing
    try:
        retrieval_start = time.time()
        result = qa_chain.invoke({"query": question})
        total_time = time.time() - start_time
        retrieval_time = time.time() - retrieval_start
        
        print(f"✓ Question answered in {total_time:.2f}s (retrieval: {retrieval_time:.2f}s)")
        return result["result"] if isinstance(result, dict) else str(result)
    except Exception as e:
        print(f"✗ Error generating answer: {e}")
        if "rate limit" in str(e).lower() or "quota" in str(e).lower():
            return "API rate limit reached. Please wait a moment and try again."
        if "fetch_k" in str(e).lower():
            return "Vector database configuration error. Please restart the application."
        return f"Error generating answer: {str(e)}"

# Create Gradio interface
with gr.Blocks(title="Research Paper RAG System") as iface:
    gr.Markdown("# 🚀 Optimized Research Paper RAG System")
    gr.Markdown("Upload PDF or DOCX research papers and ask questions about them.")
    gr.Markdown(f"**⚡ Performance Optimized**: {BATCH_SIZE}-chunk batches, {BATCH_DELAY}s delays, optimized retrieval.")

    with gr.Row():
        with gr.Column():
            file_input = gr.File(label="Upload Research Paper (PDF/DOCX)", file_types=[".pdf", ".docx"])
            upload_button = gr.Button("Process Document")
            upload_output = gr.Textbox(label="Upload Status", interactive=False)
            progress_bar = gr.Textbox(label="Processing Progress", interactive=False, visible=False)

        with gr.Column():
            question_input = gr.Textbox(label="Ask a question about the uploaded papers", lines=3)
            query_button = gr.Button("Ask Question")
            answer_output = gr.Textbox(label="Answer", lines=10, interactive=False)

    upload_button.click(process_document, inputs=file_input, outputs=upload_output)
    query_button.click(query_documents, inputs=question_input, outputs=answer_output)

if __name__ == "__main__":
    iface.launch()