Update app.py

app.py

@@ -8,6 +8,7 @@ import gradio as gr
 import plotly.graph_objects as go
 import httpx
 import logging
+from utils.llamaindex_rag import get_card_benefits_rag, initialize_rag
 
 logging.basicConfig(level=logging.INFO)
 logger = logging.getLogger(__name__)
@@ -446,6 +447,7 @@ def get_recommendation_with_agent(user_id, merchant, category, amount):
             
             yield f"❌ Invalid response: No recommendation found", None
             return
+
         
         
         card_id = recommendation.get('recommended_card', 'Unknown')
@@ -493,6 +495,26 @@ def get_recommendation_with_agent(user_id, merchant, category, amount):
         print(f"   Base rate: {base_rate}%")
         print(f"   Annual fee: ${annual_fee}")
 
+        llamaindex_context = None
+        spending_warnings = None
+        
+        if rag.enabled:  # rag was initialized at the top of the file
+            logger.info("📚 Fetching RAG context...")
+            
+            # Get card-specific context
+            llamaindex_context = rag.get_card_context(
+                card_name=card_name,  # This variable already exists in your code
+                merchant=merchant,
+                category=category
+            )
+            
+            # Get spending warnings
+            spending_warnings = rag.get_spending_warnings(
+                card_name=card_name,
+                category=category,
+                amount=amount
+            )
+
         
         # ✅ Calculate baseline comparison
         baseline_rewards = annual_spending * 0.01
@@ -574,7 +596,25 @@ def get_recommendation_with_agent(user_id, merchant, category, amount):
 
 ---
 """
-        
+
+        if llamaindex_context:
+            output += f"""
+
+---
+
+### 📚 Card Benefits & Details
+**Powered by LlamaIndex RAG**
+
+{llamaindex_context}
+"""
+    
+    # ✅ NEW: Add spending warnings
+    if spending_warnings:
+        output += f"""
+
+### ⚠️ Important to Know
+{spending_warnings}
+"""
         if alternatives:
             output += "\n### 🔄 Alternative Options\n\n"
             output += "| Rank | Card | Rewards | Rate | Why Ranked Here? |\n"
@@ -656,6 +696,12 @@ def get_recommendation_with_agent(user_id, merchant, category, amount):
         
         # ✅ Create chart
         chart = create_agent_recommendation_chart_enhanced(recommendation)
+        if llamaindex_context or spending_warnings:
+        output += f"""
+---
+
+*🤖 Enhanced with AI-powered knowledge retrieval using LlamaIndex + OpenAI Embeddings*
+"""
         
         yield output, chart
         
@@ -729,6 +775,13 @@ client = RewardPilotClient(config.ORCHESTRATOR_URL)
 # llm = get_llm_explainer()
 gemini = get_gemini_explainer()
 
+logger.info("🚀 Initializing LlamaIndex RAG...")
+card_rag = initialize_rag()
+if card_rag.enabled:
+    logger.info("✅ RAG system ready")
+else:
+    logger.warning("⚠️ RAG system not available")
+    
 def get_recommendation(
     user_id: str,
     merchant: str,
@@ -2971,11 +3024,52 @@ with gr.Blocks(
         
             
             def respond(message, chat_history, user_id, use_voice=False, voice_name="Rachel", voice_speed=1.0):
-                """Enhanced chat with OpenAI GPT-4 and optional voice output"""
+                """Enhanced chat with OpenAI GPT-4, LlamaIndex RAG, and optional voice output"""
                 
                 if not message.strip():
                     return "", chat_history, None
                 
+                # ✅ NEW: Check if question is about card benefits (RAG integration)
+                rag = get_card_benefits_rag()
+                rag_context = None
+                
+                if rag.enabled:
+                    # Detect if question is about specific card
+                    card_keywords = ["amex", "gold", "chase", "sapphire", "reserve", "freedom", "unlimited", 
+                                   "citi", "double cash", "discover", "card", "benefit", "reward", "point",
+                                   "cashback", "annual fee", "cap", "limit", "grocery", "dining", "travel"]
+                    
+                    message_lower = message.lower()
+                    if any(keyword in message_lower for keyword in card_keywords):
+                        logger.info("📚 Detected card-specific question, querying RAG...")
+                        
+                        # Extract card name (simple heuristic)
+                        card_name = None
+                        if "sapphire reserve" in message_lower or "csr" in message_lower:
+                            card_name = "Chase Sapphire Reserve"
+                        elif "freedom unlimited" in message_lower or "cfu" in message_lower:
+                            card_name = "Chase Freedom Unlimited"
+                        elif "double cash" in message_lower:
+                            card_name = "Citi Double Cash"
+                        elif "discover" in message_lower:
+                            card_name = "Discover it"
+                        elif "amex gold" in message_lower or "american express gold" in message_lower:
+                            card_name = "Amex Gold"
+                        elif "gold" in message_lower and ("amex" in message_lower or "american express" in message_lower):
+                            card_name = "Amex Gold"
+                        
+                        # Query RAG if card was identified
+                        if card_name:
+                            try:
+                                rag_context = rag.query_benefits(card_name, message)
+                                if rag_context:
+                                    logger.info(f"✅ RAG context retrieved: {len(rag_context)} chars for {card_name}")
+                                else:
+                                    logger.warning(f"⚠️ RAG returned no context for {card_name}")
+                            except Exception as e:
+                                logger.error(f"❌ RAG query failed: {e}")
+                                rag_context = None
+                
                 # Get user context (your existing logic)
                 user_context = {}
                 try:
@@ -3010,11 +3104,8 @@ with gr.Blocks(
                     }
                 ]
                 
-                # Build messages (your existing logic)
-                messages = [
-                    {
-                        "role": "system",
-                        "content": f"""You are CardWise AI, an expert credit card rewards optimizer.
+                # Build messages (your existing logic with RAG enhancement)
+                system_content = f"""You are CardWise AI, an expert credit card rewards optimizer.
             
             User Context:
             - User ID: {user_context.get('user_id', 'Unknown')}
@@ -3024,6 +3115,21 @@ with gr.Blocks(
             
             When voice mode is enabled, keep responses concise and conversational (under 200 words).
             Be helpful, actionable, and friendly."""
+                
+                # ✅ NEW: Add RAG context if available
+                if rag_context:
+                    system_content += f"""
+            
+            📚 KNOWLEDGE BASE CONTEXT:
+            {rag_context}
+            
+            Use this information to provide accurate, detailed answers about card benefits.
+            Always cite specific details from the knowledge base when relevant."""
+                
+                messages = [
+                    {
+                        "role": "system",
+                        "content": system_content
                     }
                 ]
                 
@@ -3088,6 +3194,10 @@ with gr.Blocks(
                     else:
                         bot_response = response_message.content
                     
+                    # ✅ NEW: Add RAG attribution if context was used
+                    if rag_context:
+                        bot_response += "\n\n*📚 Enhanced with knowledge base using LlamaIndex*"
+                    
                     print(f"✅ GPT-4 response generated")
                     
                     # Generate voice if enabled
@@ -3095,8 +3205,12 @@ with gr.Blocks(
                     if use_voice and voice_assistant.enabled:
                         try:
                             logger.info(f"🎤 Generating voice with {voice_name}")
+                            
+                            # ✅ IMPROVED: Clean response for voice (remove RAG attribution)
+                            voice_text = bot_response.replace("*📚 Enhanced with knowledge base using LlamaIndex*", "").strip()
+                            
                             audio_bytes = voice_assistant.text_to_speech(
-                                text=bot_response,
+                                text=voice_text,
                                 voice_name=voice_name
                             )
                             
@@ -3124,7 +3238,6 @@ with gr.Blocks(
                 chat_history.append((message, bot_response))
                 return "", chat_history, audio_output
 
-
             
             
             def generate_fallback_response(message: str, user_context: dict) -> str:
@@ -3644,7 +3757,225 @@ with gr.Blocks(
                 outputs=[receipt_output, receipt_chart]
             )
 
+
+                # ==================== TAB: CARD KNOWLEDGE BASE (RAG) ====================
+        with gr.Tab("📚 Knowledge Base"):
+            rag = get_card_benefits_rag()
             
+            if rag.enabled:
+                gr.HTML("""
+                <div style="background: linear-gradient(135deg, #f093fb 0%, #f5576c 100%); 
+                            padding: 30px; border-radius: 16px; color: white; margin-bottom: 25px;">
+                    <h2 style="margin: 0 0 15px 0;">📚 AI-Powered Card Knowledge Base</h2>
+                    <p style="margin: 0; font-size: 17px; line-height: 1.7;">
+                        Search our comprehensive credit card database using <strong>LlamaIndex RAG</strong> 
+                        powered by OpenAI embeddings and GPT-4.
+                    </p>
+                    <div style="background: rgba(255,255,255,0.15); padding: 15px; 
+                                border-radius: 10px; margin-top: 15px;">
+                        <p style="margin: 0; font-size: 15px;">
+                            🔍 <strong>Semantic Search:</strong> Ask natural language questions<br>
+                            🎯 <strong>Accurate Answers:</strong> Powered by vector embeddings<br>
+                            ⚡ <strong>Real-time:</strong> Instant retrieval from knowledge base
+                        </p>
+                    </div>
+                </div>
+                """)
+                
+                gr.Markdown("## 🔍 Search Card Benefits")
+                
+                with gr.Row():
+                    kb_card = gr.Dropdown(
+                        choices=[
+                            "Amex Gold",
+                            "American Express Gold",
+                            "Chase Sapphire Reserve",
+                            "Chase Freedom Unlimited",
+                            "Citi Double Cash",
+                            "Discover it"
+                        ],
+                        label="💳 Select Card",
+                        value="Amex Gold",
+                        scale=1
+                    )
+                    kb_query = gr.Textbox(
+                        label="❓ Your Question",
+                        placeholder="e.g., Does this card work at Costco for groceries?",
+                        scale=3
+                    )
+                
+                kb_search_btn = gr.Button("🔍 Search Knowledge Base", variant="primary", size="lg")
+                
+                kb_result = gr.Markdown(value="*Enter a question and click Search*")
+                
+                def search_knowledge_base(card, question):
+                    """Search card benefits using LlamaIndex RAG"""
+                    if not question or not question.strip():
+                        return "⚠️ Please enter a question"
+                    
+                    rag = get_card_benefits_rag()
+                    
+                    if not rag.enabled:
+                        return """
+## ⚠️ RAG Not Available
+
+The LlamaIndex RAG system is not currently enabled. This could be due to:
+
+1. Missing dependencies (llama-index not installed)
+2. No OpenAI API key configured
+3. No card benefit documents in data/card_benefits/
+
+Please check the logs for more details.
+"""
+                    
+                    try:
+                        logger.info(f"🔍 Knowledge base search: {card} - {question}")
+                        result = rag.query_benefits(card, question)
+                        
+                        if result:
+                            return f"""
+## 🎯 Answer for {card}
+
+{result}
+
+---
+
+### 📊 Search Details
+- **Card:** {card}
+- **Question:** {question}
+- **Source:** LlamaIndex RAG with GPT-4
+- **Embeddings:** OpenAI text-embedding-3-small
+
+*💡 This answer was retrieved using semantic search across our card benefits knowledge base.*
+"""
+                        else:
+                            return f"""
+## ❌ No Results Found
+
+Could not find relevant information about **{card}** for your question:
+
+> {question}
+
+### 💡 Suggestions:
+- Try rephrasing your question
+- Check if the card name is correct
+- Ask about specific features (earning rates, caps, exclusions)
+"""
+                    
+                    except Exception as e:
+                        logger.error(f"❌ Knowledge base search failed: {e}")
+                        return f"## ❌ Search Error\n\nAn error occurred: {str(e)}"
+                
+                kb_search_btn.click(
+                    fn=search_knowledge_base,
+                    inputs=[kb_card, kb_query],
+                    outputs=[kb_result]
+                )
+                
+                gr.Markdown("---")
+                gr.Markdown("### 💡 Example Questions")
+                
+                gr.Examples(
+                    examples=[
+                        ["Amex Gold", "Does this card work at Costco for groceries?"],
+                        ["Amex Gold", "What's the annual spending cap on grocery purchases?"],
+                        ["Chase Sapphire Reserve", "What travel benefits does this card offer?"],
+                        ["Chase Sapphire Reserve", "Does Uber count as travel for earning points?"],
+                        ["Chase Freedom Unlimited", "What's the earning rate on dining?"],
+                        ["Citi Double Cash", "How does the 2% cash back work?"],
+                        ["Discover it", "What are the rotating categories this quarter?"],
+                    ],
+                    inputs=[kb_card, kb_query],
+                    label="Try these questions"
+                )
+                
+                # Card comparison feature
+                gr.Markdown("---")
+                gr.Markdown("## ⚖️ Compare Cards")
+                
+                with gr.Row():
+                    compare_card1 = gr.Dropdown(
+                        choices=["Amex Gold", "Chase Sapphire Reserve", "Chase Freedom Unlimited", 
+                                "Citi Double Cash", "Discover it"],
+                        label="Card 1",
+                        value="Amex Gold"
+                    )
+                    compare_card2 = gr.Dropdown(
+                        choices=["Amex Gold", "Chase Sapphire Reserve", "Chase Freedom Unlimited", 
+                                "Citi Double Cash", "Discover it"],
+                        label="Card 2",
+                        value="Chase Sapphire Reserve"
+                    )
+                    compare_category = gr.Dropdown(
+                        choices=["Dining", "Groceries", "Travel", "Gas", "General Spending"],
+                        label="Category",
+                        value="Dining"
+                    )
+                
+                compare_btn = gr.Button("⚖️ Compare Cards", variant="secondary", size="lg")
+                compare_result = gr.Markdown(value="*Select cards and click Compare*")
+                
+                def compare_cards_ui(card1, card2, category):
+                    """Compare two cards for a specific category"""
+                    if card1 == card2:
+                        return "⚠️ Please select two different cards"
+                    
+                    rag = get_card_benefits_rag()
+                    
+                    if not rag.enabled:
+                        return "⚠️ RAG not available"
+                    
+                    try:
+                        result = rag.compare_cards(card1, card2, category)
+                        
+                        if result:
+                            return f"""
+## ⚖️ Comparison: {card1} vs {card2}
+
+### Category: {category}
+
+{result}
+
+---
+
+*📚 Powered by LlamaIndex RAG*
+"""
+                        else:
+                            return "❌ Could not generate comparison"
+                    
+                    except Exception as e:
+                        logger.error(f"❌ Comparison failed: {e}")
+                        return f"❌ Error: {str(e)}"
+                
+                compare_btn.click(
+                    fn=compare_cards_ui,
+                    inputs=[compare_card1, compare_card2, compare_category],
+                    outputs=[compare_result]
+                )
+                
+            else:
+                # RAG not available
+                gr.HTML("""
+                <div style="background: linear-gradient(135deg, #ff6b6b 0%, #ee5a6f 100%); 
+                            padding: 30px; border-radius: 16px; color: white; margin-bottom: 25px;">
+                    <h2 style="margin: 0 0 15px 0;">⚠️ Knowledge Base Not Available</h2>
+                    <p style="margin: 0; font-size: 17px;">
+                        The LlamaIndex RAG system requires additional setup.
+                    </p>
+                </div>
+                """)
+                
+                gr.Markdown("""
+## 🔧 Setup Required
+
+To enable the Knowledge Base:
+
+1. Install dependencies: `pip install llama-index`
+2. Add OpenAI API key
+3. Create card benefit documents in `data/card_benefits/`
+4. Restart the application
+""")
+        
         # ==================== TAB 6: RESOURCES (About + Agent Insight + API Docs) ====================
         with gr.Tab("ℹ️ Resources"):
             with gr.Tabs():