Update app.py

app.py

@@ -332,40 +332,30 @@ def generate_llm_answer(
         return None
     
     # Sort and take top 8
-    scored_docs.sort(key=lambda x: x[1], reverse=True)
-    top_docs = [doc[0] for doc in scored_docs[:8]]
-    
-    # Build context - keep it SHORT to stay under 512 tokens
-    context_parts = []
-    for doc in top_docs[:5]:  # Only use top 5 docs
-        content = doc.page_content.strip()
-        # Keep each doc snippet under 150 chars
-        if len(content) > 150:
-            content = content[:150] + "..."
-        context_parts.append(content)
-    
-    context_text = "\n\n".join(context_parts)
-    
-    # Progressive parameters - optimized for SPEED (shorter = faster)
+    # Optimized parameters for 2-attempt strategy
     if attempt == 1:
-        temperature = 0.7
-        max_new_tokens = 250  # Faster generation
-        top_p = 0.9
+        temperature = 0.75
+        max_tokens = 350
+        top_p = 0.92
+        repetition_penalty = 1.15
+    else:  # attempt == 2
+        temperature = 0.85
+        max_tokens = 450
+        top_p = 0.94
         repetition_penalty = 1.2
-    elif attempt == 2:
         temperature = 0.75
         max_new_tokens = 300
         top_p = 0.92
         repetition_penalty = 1.25
-    elif attempt == 3:
-        temperature = 0.8
-        max_new_tokens = 350
-        top_p = 0.94
-        repetition_penalty = 1.3
-    else:
-        temperature = 0.85
-        max_new_tokens = 400
-        top_p = 0.95
+    # T5 format - simple and effective for good answers
+    user_prompt = f"""Answer this fashion question with detailed, specific advice using the context provided.
+
+Question: {query}
+
+Fashion Context:
+{context_text[:1500]}
+
+Provide a complete, detailed answer (150-250 words):"""
         repetition_penalty = 1.35
     
     # Create COMPACT T5 prompt to stay under 512 tokens (critical!)
@@ -429,36 +419,10 @@ Fashion Answer:"""
         return response
         
     except Exception as e:
-        logger.error(f"  ✗ Generation error: {e}")
-        return None
-
-# ============================================================================
-# MAIN RAG FUNCTION
-# ============================================================================
-
-def generate_answer_langchain(
-    query: str,
-    vectorstore,
-    llm_client
-) -> str:
-    """
-    Main RAG pipeline: Retrieve → Generate (no fallback)
-    """
-    logger.info(f"\n{'='*80}")
-    logger.info(f"Processing query: '{query}'")
-    logger.info(f"{'='*80}")
-    
-    # Step 1: Retrieve documents
-    retrieved_docs, confidence = retrieve_knowledge_langchain(
-        query,
-        vectorstore,
-        top_k=CONFIG["top_k"]
-    )
-    
     if not retrieved_docs:
         return "I couldn't find relevant information to answer your question."
     
-    # Step 2: Try LLM generation (2 attempts for speed)
+    # Step 2: Try LLM generation (2 fast attempts for efficiency)
     llm_answer = None
     for attempt in range(1, 3):
         logger.info(f"\n  🤖 LLM Generation Attempt {attempt}/2")
@@ -475,12 +439,12 @@ def generate_answer_langchain(
         logger.error(f"  ✗ All 2 LLM attempts failed")
         return "I apologize, but I'm having trouble generating a response. Please try rephrasing your question or ask something else."
     
-    return llm_answer
-
-# ============================================================================
-# GRADIO INTERFACE
-# ============================================================================
-
+    return llm_answeronfidence = retrieve_knowledge_langchain(
+        query,
+        vectorstore,
+        top_k=CONFIG["top_k"]
+    )
+    
 def fashion_chatbot(message: str, history: List[List[str]]):
     """
     Chatbot function for Gradio interface with streaming
@@ -490,11 +454,37 @@ def fashion_chatbot(message: str, history: List[List[str]]):
             yield "Please ask a fashion-related question!"
             return
         
-        # ANTI-HALLUCINATION: Validate if question is fashion-related
-        query_lower = message.strip().lower()
-        fashion_keywords = [
-            'wear', 'outfit', 'dress', 'style', 'fashion', 'clothing', 'clothes',
-            'color', 'match', 'look', 'shirt', 'pants', 'shoes', 'accessory',
+        # Show searching indicator
+        yield "🔍 Searching fashion knowledge..."d successfully")
+            break
+        else:
+            logger.warning(f"  → Attempt {attempt}/2 failed, retrying...")
+    
+    # Step 3: If all attempts fail, return error
+    if not llm_answer:
+        logger.error(f"  ✗ All 2 LLM attempts failed")
+        return "I apologize, but I'm having trouble generating a response. Please try rephrasing your question or ask something else."
+    
+    return llm_answer
+
+# ============================================================================
+# GRADIO INTERFACE
+# ============================================================================
+
+        # Generate answer with 2 fast attempts
+        llm_answer = None
+        for attempt in range(1, 3):
+            logger.info(f"\n  🤖 LLM Generation Attempt {attempt}/2")
+            llm_answer = generate_llm_answer(message.strip(), retrieved_docs, llm_client, attempt)
+            
+            if llm_answer:
+                break
+        
+        # If LLM fails, show error
+        if not llm_answer:
+            logger.error(f"  ✗ All LLM attempts failed")
+            yield "I apologize, but I'm having trouble generating a response. Please try rephrasing your question."
+            return', 'match', 'look', 'shirt', 'pants', 'shoes', 'accessory',
             'wardrobe', 'fit', 'fabric', 'pattern', 'casual', 'formal', 'seasonal',
             'wedding', 'meeting', 'interview', 'date', 'party', 'jeans', 'suit',
             'skirt', 'jacket', 'coat', 'sweater', 'blouse', 'tie', 'scarf', 'boots',