Update app.py

app.py

@@ -330,14 +330,16 @@ class ProductionRAGSystem:
         # Always generate extracted answer
         extracted_answer = self.extract_direct_answer(query, best_result['content'])
         
-        # Try AI answer with minimal tokens
+        # Try AI answer with minimal tokens - send only extracted answer, not full chunks
         ai_answer = None
         openrouter_key = os.environ.get("OPENROUTER_API_KEY")
         
         if openrouter_key:
-            # Use much shorter context and fewer tokens
-            context = search_results[0]['content'][:200]  # Reduced from 500
-            prompt = f"Q: {query}\nA:"  # Much shorter prompt
+            # Use the extracted answer as context instead of raw chunks
+            extracted_answer = self.extract_direct_answer(query, best_result['content'])
+            
+            # Super minimal prompt with just the extracted info
+            prompt = f"Improve: {extracted_answer}"  # Very short prompt
             
             try:
                 response = requests.post(
@@ -351,14 +353,16 @@ class ProductionRAGSystem:
                     json={
                         "model": "openai/gpt-3.5-turbo",
                         "messages": [{"role": "user", "content": prompt}],
-                        "max_tokens": 30,  # Reduced from 100
+                        "max_tokens": 25,  # Very small
                         "temperature": 0.1
                     },
                     timeout=10
                 )
                 
                 if response.status_code == 200:
-                    ai_answer = response.json()['choices'][0]['message']['content'].strip()
+                    ai_response = response.json()['choices'][0]['message']['content'].strip()
+                    # Use AI response if it's actually better, otherwise stick with extracted
+                    ai_answer = ai_response if len(ai_response) > 10 else extracted_answer
                 elif response.status_code == 402:
                     st.error("💳 OpenRouter credits exhausted. Using extracted answers only.")
                 else:
@@ -366,14 +370,23 @@ class ProductionRAGSystem:
                     
             except Exception as e:
                 st.error(f"API Exception: {str(e)}")
-        
-        return {
-            'ai_answer': ai_answer,
-            'extracted_answer': extracted_answer,
-            'sources': sources,
-            'confidence': avg_confidence,
-            'has_both': ai_answer is not None
-        }
+            
+            return {
+                'ai_answer': ai_answer,
+                'extracted_answer': extracted_answer,
+                'sources': sources,
+                'confidence': avg_confidence,
+                'has_both': ai_answer is not None
+            }
+        else:
+            # No API key - just return extracted answer
+            return {
+                'ai_answer': None,
+                'extracted_answer': extracted_answer,
+                'sources': sources,
+                'confidence': avg_confidence,
+                'has_both': False
+            }
 
 def get_general_ai_response(query):
     """Get AI response for general questions with minimal token usage"""
@@ -676,8 +689,8 @@ if prompt := st.chat_input("Ask questions about your documents..."):
             # Search documents first
             search_results = rag_system.search(prompt, n_results=3)
             
-            # Check if we found relevant documents (confidence > 0.05)
-            if search_results and search_results[0]['similarity'] > 0.05:
+            # Check if we found relevant documents (much lower threshold)
+            if search_results and search_results[0]['similarity'] > 0.01:  # Very low threshold
                 # Generate document-based answer
                 result = rag_system.generate_answer(prompt, search_results)