app.py

import gradio as gr
import tempfile
import os
import logging
from typing import List, Dict, Any, Optional, Tuple
import time
from datetime import datetime
import json

# Configure logging
logging.basicConfig(level=logging.INFO)
logger = logging.getLogger(__name__)

# Import our modules
from document_processor_hf import DocumentProcessor
from inference_client import GemmaInferenceClient

# Global variables for maintaining state
processor = DocumentProcessor()
inference_client = GemmaInferenceClient()
current_document = None
chat_history = []
processing_stats = {}

def process_document(file, use_smart_processing, use_prf, use_variants, use_reranking, progress=gr.Progress()):
    """Process uploaded document"""
    global current_document, processing_stats
    
    if file is None:
        return "❌ Please upload a document first.", "", ""
    
    try:
        progress(0.1, desc="📄 Reading document...")
        
        # Process the document
        result = processor.process_document(
            file.name, 
            use_smart_processing=use_smart_processing
        )
        
        progress(0.7, desc="🧠 Extracting entities...")
        
        if result['success']:
            current_document = {
                'name': os.path.basename(file.name),
                'type': result.get('document_type', 'general'),
                'chunks': len(result['chunks']),
                'entities': len(result['entities']),
                'suggestions': result.get('suggestions', [])
            }
            
            processing_stats = {
                'document_type': result.get('document_type', 'general'),
                'entities_found': len(result['entities']),
                'chunks_created': len(result['chunks']),
                'text_length': result.get('text_length', 0)
            }
            
            progress(1.0, desc="✅ Processing complete!")
            
            # Create document info display
            doc_info = f"""
## 📊 Document Information

**📄 File:** {current_document['name']}  
**📝 Type:** {current_document['type'].title()}  
**📚 Chunks:** {current_document['chunks']}  
**🏷️ Entities:** {current_document['entities']}  

### 🎯 Active Enhancements:
{f"🔍 **Pseudo Relevance Feedback**: {'✅ Enabled' if use_prf else '❌ Disabled'}" if use_smart_processing else ""}  
{f"📝 **Query Variants**: {'✅ Enabled' if use_variants else '❌ Disabled'}" if use_smart_processing else ""}  
{f"🎯 **Cross-Encoder Reranking**: {'✅ Enabled' if use_reranking else '❌ Disabled'}" if use_smart_processing else ""}
"""
            
            # Create suggested questions
            suggestions_html = ""
            if current_document['suggestions']:
                suggestions_html = "### 💡 Suggested Questions:\n"
                for i, suggestion in enumerate(current_document['suggestions'][:5]):
                    suggestions_html += f"{i+1}. {suggestion}\n"
            
            success_msg = f"✅ **Document processed successfully!** Ready for questions."
            
            return success_msg, doc_info, suggestions_html
            
        else:
            error_msg = f"❌ **Processing failed:** {result.get('error', 'Unknown error')}"
            return error_msg, "", ""
            
    except Exception as e:
        logger.error(f"Document processing error: {e}")
        error_msg = f"❌ **Error:** {str(e)}"
        return error_msg, "", ""

def chat_with_document(message, history, temperature, max_tokens, top_k, use_prf, use_variants, use_reranking):
    """Chat with the processed document"""
    global current_document
    
    if not current_document:
        history.append([message, "❌ Please upload and process a document first."])
        return history, ""
    
    if not message.strip():
        return history, ""
    
    try:
        # Add thinking message
        history.append([message, "🤔 Thinking..."])
        yield history, ""
        
        # Query the document
        start_time = time.time()
        
        context_result = processor.query_document(
            message,
            top_k=min(top_k, 3),  # Limit context for memory
            use_smart_retrieval=True,
            use_prf=use_prf,
            use_variants=use_variants,
            use_reranking=use_reranking
        )
        
        # Generate response with memory-efficient settings
        response_result = inference_client.generate_response(
            query=message,
            context=context_result['context'][:2000],  # Limit context length
            temperature=temperature,
            max_tokens=min(max_tokens, 256)  # Limit response length for memory
        )
        
        query_time = time.time() - start_time
        
        # Format response with enhancements info
        response = response_result['response']
        
        # Add enhancement information
        enhancements = []
        if use_prf:
            enhancements.append("🔍 PRF")
        if use_variants:
            enhancements.append("📝 Variants")
        if use_reranking:
            enhancements.append("🎯 Reranking")
        
        if enhancements:
            response += f"\n\n*Enhanced with: {' | '.join(enhancements)} | ⚡ {query_time:.2f}s*"
        
        # Clear memory after each response
        inference_client.clear_cache()
        
        # Update history with final response
        history[-1] = [message, response]
        yield history, ""
        
    except Exception as e:
        logger.error(f"Chat error: {e}")
        error_response = f"❌ Sorry, I encountered an error: {str(e)}"
        history[-1] = [message, error_response]
        yield history, ""

def use_suggested_question(question_text, history):
    """Use a suggested question"""
    if question_text and current_document:
        return question_text, history
    return "", history

def clear_chat():
    """Clear chat history"""
    return []

def get_example_files():
    """Get example file information"""
    examples = """
### 📋 Try these document types:

**📄 Resumes/CVs**: Upload a resume to ask "Whose resume is this?" or "What are their skills?"

**📊 Reports**: Upload a business report to ask "What are the key findings?" or "What methodology was used?"

**📋 Contracts**: Upload a contract to ask "What are the main terms?" or "Who are the parties involved?"

**📚 Academic Papers**: Upload a research paper to ask "What is the research question?" or "What are the results?"

**🖼️ Images**: Upload screenshots or scanned documents with text for OCR processing.
"""
    return examples

# Create the Gradio interface
with gr.Blocks(
    title="Document Chat with Gemma 3",
    theme=gr.themes.Soft(
        primary_hue="blue",
        secondary_hue="purple",
        neutral_hue="slate"
    ),
    css="""
    .gradio-container {
        max-width: 1200px !important;
    }
    .main-header {
        text-align: center;
        background: linear-gradient(90deg, #667eea 0%, #764ba2 100%);
        color: white;
        padding: 2rem;
        border-radius: 10px;
        margin-bottom: 2rem;
    }
    .enhancement-badge {
        display: inline-block;
        padding: 0.25rem 0.5rem;
        margin: 0.1rem;
        border-radius: 15px;
        font-size: 0.8rem;
        font-weight: bold;
        background-color: #e3f2fd;
        color: #1976d2;
    }
    .document-info {
        background-color: #f8f9fa;
        padding: 1rem;
        border-radius: 8px;
        border-left: 4px solid #667eea;
    }
    """
) as demo:
    
    # Header
    gr.HTML("""
    <div class="main-header">
        <h1>📄 Document Chat with Gemma 3</h1>
        <p>Upload documents and chat with advanced RAG capabilities powered by Gemma 3</p>
        <p><strong>🚀 Features:</strong> Smart Entity Extraction | Document Type Detection | Query Enhancement | Context-Aware Responses</p>
    </div>
    """)
    
    with gr.Row():
        # Left column - Document Upload and Configuration
        with gr.Column(scale=1):
            gr.Markdown("## 📤 Upload & Configure")
            
            # File upload
            file_input = gr.File(
                label="📄 Upload Document",
                file_types=[".pdf", ".txt", ".md", ".csv", ".docx", ".jpg", ".jpeg", ".png"],
                type="filepath"
            )
            
            # Processing options
            with gr.Group():
                gr.Markdown("### 🛠️ Processing Options")
                use_smart_processing = gr.Checkbox(
                    label="🧠 Smart Processing", 
                    value=True,
                    info="Enable entity extraction and document type detection"
                )
                use_prf = gr.Checkbox(
                    label="🔍 Pseudo Relevance Feedback", 
                    value=True,
                    info="Expand queries using relevant document terms"
                )
                use_variants = gr.Checkbox(
                    label="📝 Query Variants", 
                    value=True,
                    info="Generate multiple query reformulations"
                )
                use_reranking = gr.Checkbox(
                    label="🎯 Cross-Encoder Reranking", 
                    value=True,
                    info="Rerank results using advanced models"
                )
            
            # Model parameters
            with gr.Group():
                gr.Markdown("### ⚙️ Model Parameters")
                temperature = gr.Slider(
                    minimum=0.0, maximum=1.0, value=0.3, step=0.1,
                    label="🌡️ Temperature",
                    info="Controls response creativity"
                )
                max_tokens = gr.Slider(
                    minimum=64, maximum=256, value=128, step=32,
                    label="📏 Max Tokens",
                    info="Maximum response length (limited for memory)"
                )
                top_k = gr.Slider(
                    minimum=3, maximum=10, value=5, step=1,
                    label="📚 Context Chunks",
                    info="Number of document chunks to retrieve"
                )
            
            # Process button
            process_btn = gr.Button(
                "🚀 Process Document", 
                variant="primary",
                size="lg"
            )
            
            # Processing status
            process_status = gr.Markdown("")
            
        # Right column - Chat Interface
        with gr.Column(scale=2):
            gr.Markdown("## 💬 Chat with Your Document")
            
            # Document info display
            doc_info_display = gr.Markdown("", visible=False)
            
            # Suggested questions
            suggested_questions_display = gr.Markdown("", visible=False)
            
            # Chat interface
            chatbot = gr.Chatbot(
                height=400,
                show_label=False,
                container=True,
                show_copy_button=True
            )
            
            with gr.Row():
                msg_input = gr.Textbox(
                    placeholder="Ask a question about your document...",
                    show_label=False,
                    scale=4,
                    container=False
                )
                send_btn = gr.Button("Send", variant="primary", scale=1)
                clear_btn = gr.Button("Clear", variant="secondary", scale=1)
    
    # Bottom section - Examples and tips
    with gr.Row():
        with gr.Column():
            gr.Markdown(get_example_files())
        
        with gr.Column():
            gr.Markdown("""
            ### 🎯 Pro Tips:
            
            **📋 For Resumes**: Ask "Whose resume is this?" or "What are their technical skills?"
            
            **📊 For Reports**: Ask "What are the key findings?" or "What methodology was used?"
            
            **📜 For Contracts**: Ask "What are the main terms?" or "Who are the parties involved?"
            
            **🔍 Advanced Queries**: The system understands context and can answer complex questions about relationships, timelines, and document structure.
            
            **⚡ Performance**: First query may take longer as models load. Subsequent queries are faster.
            """)
    
    # Event handlers
    def process_and_update(file, smart, prf, variants, rerank):
        """Process document and update displays"""
        status, doc_info, suggestions = process_document(file, smart, prf, variants, rerank)
        
        # Show/hide info panels based on success
        doc_info_visible = "✅" in status
        suggestions_visible = bool(suggestions.strip()) if suggestions else False
        
        return (
            status,
            gr.update(value=doc_info, visible=doc_info_visible),
            gr.update(value=suggestions, visible=suggestions_visible)
        )
    
    # Connect event handlers
    process_btn.click(
        fn=process_and_update,
        inputs=[file_input, use_smart_processing, use_prf, use_variants, use_reranking],
        outputs=[process_status, doc_info_display, suggested_questions_display]
    )
    
    # Chat functionality
    msg_input.submit(
        fn=chat_with_document,
        inputs=[msg_input, chatbot, temperature, max_tokens, top_k, use_prf, use_variants, use_reranking],
        outputs=[chatbot, msg_input]
    )
    
    send_btn.click(
        fn=chat_with_document,
        inputs=[msg_input, chatbot, temperature, max_tokens, top_k, use_prf, use_variants, use_reranking],
        outputs=[chatbot, msg_input]
    )
    
    clear_btn.click(
        fn=clear_chat,
        outputs=[chatbot]
    )

# Launch configuration
if __name__ == "__main__":
    demo.launch(
        server_name="0.0.0.0",
        server_port=7860,
        share=False,
        show_api=False,
        show_error=True
    )