Fix to handle user queries already having a credit card

recommender/retrieval_ranking.py

@@ -14,46 +14,87 @@ def convert_to_direct_query_gradio(user_query,preferences):
     preferences_text = ""
     if preferences:
         preferences_text = "User selected preferences: " + ", ".join(preferences) + "."
-    # Prompt with examples for indirect-to-direct conversion
+
+    card_list_str = ", ".join(all_card_names)
+    feature_list_str = ", ".join(feature_list)
+
     prompt = f"""
-        You are an AI assistant that refines user queries to make them optimized for information retrieval while keeping the original intent.  
-    
-        Instructions:  
-        - Identify the main focus from the query.  
-        - Reformat the query in a structured way for better retrieval.  
-        - Ensure the most important feature appears first.  
-        - Use precise keywords that match credit card benefits.  
-        - Include the preferences also in the final query.  
-        - Do NOT introduce new benefits not mentioned by the user.  
-        - If the user uses vague terms like "vacation", interpret it as travel-related benefits including: airport lounge access, international/domestic travel, hotel benefits, forex waiver.  
-        - If the query includes terms like "beginner", "entry-level", or "low credit score", include essential features such as cashback, reward points, and basic offers.  
-        - If the query is empty, generate a useful retrieval-focused query based solely on preferences.  
-        
-        Examples:  
-    
+        You are an AI assistant that refines user queries to make them optimized for credit card information retrieval while strictly preserving the user's original intent.
+        Your task is to restructure queries in a way that enables accurate and relevant credit card recommendations.
+        
+        Instructions:
+        - Identify the main intent or feature of the query (e.g., cashback, lounge access, air miles).
+        - Format the query clearly and concisely, with the primary benefit mentioned first.
+        - Use direct and specific keywords that match real credit card benefits.
+        - Respect and incorporate any stated user preferences.
+        - Do NOT introduce any benefits not explicitly requested by the user.
+        - Retain and highlight any mention of a specific bank or card brand if provided.
+        - If the user uses vague terms like "vacation", interpret it as related to travel benefits, including: airport lounge access, hotel privileges, travel offers, international usage, and forex fee waivers.
+        - If the user mentions "beginner", "entry-level", or "low credit score", focus on basic features like cashback, reward points, and easy approvals.
+        - If the query is empty, generate a structured, useful query based entirely on preferences.
+        
+        **Card & Feature Awareness:**
+        Only exclude cards if the user clearly states they already have them. Match this against the known cards: [{card_list_str}].
+        
+        Do not guess or assume exclusions based on the features the user mentions — only exclude when there is a direct statement like:
+        
+        "I already have <card name>"
+        
+        "I use <card name>"
+        
+        "I have a card with <specific benefit>"
+        
+        If the user does not explicitly mention owning a specific card, leave the excluded_cards array empty.
+        - If the user already has a card covering certain benefits (e.g., airline miles), do NOT recommend similar benefits again.
+        - Instead, refocus the query on alternative features that support the same intent. For example:
+          - If the user already has airline miles but asks for "frequent flyer" cards, suggest lounge access, hotel perks, or concierge services instead.
+        - Use the following set of all possible benefits to guide replacements: [{feature_list_str}].
+        - Do NOT remove or alter any request for a specific bank or product if explicitly mentioned by the user.
+        
+        Output:
+        - Return a JSON object with:
+          {{
+            "optimized_query": "Category: <Primary Feature> | <Refined, clean version of the user's query using precise benefit terms. Avoid mentioning excluded cards.>",
+            "excluded_cards": ["<card1>", "<card2>"]
+          }}
+        - Do NOT include phrases like “Exclude: <Card Name>” in the `optimized_query`.
+        
+        **Examples:**
+        
         Example 1  
         - User Query: "I drive a lot for work and want a credit card with good fuel rewards and travel perks."  
-        - Optimized Query: "Category: Fuel Rewards | Best credit cards for high fuel spending with maximum rewards & fuel surcharge waiver. Travel perks preferred but secondary."  
-    
+        - Optimized Query: "Category: Fuel Rewards | Best credit cards for high fuel spending with maximum rewards & fuel surcharge waiver. Travel perks preferred but secondary."
+        
         Example 2  
         - User Query: "I mostly shop online and want a card that gives high cashback on e-commerce purchases. Food delivery perks would be nice."  
-        - Optimized Query: "Category: Online Shopping | Credit cards with best cashback on e-commerce platforms like Amazon, Flipkart. Food delivery benefits secondary."  
-    
+        - Optimized Query: "Category: Online Shopping | Credit cards with best cashback on e-commerce platforms like Amazon, Flipkart. Food delivery benefits secondary."
+        
         Example 3  
         - User Query: "I eat out a lot and also order food from Swiggy/Zomato. I want the best dining discounts and food delivery cashback."  
-        - Optimized Query: "Category: Dining & Food Delivery | Top credit cards offering the best dining discounts at restaurants and cashback on Swiggy/Zomato orders."  
-    
-        Now, optimize the following preferences and user query:  
-    
+        - Optimized Query: "Category: Dining & Food Delivery | Top credit cards offering the best dining discounts at restaurants and cashback on Swiggy/Zomato orders."
+        
+        Example 4  
+        - User Query: "I already have a credit card with airline miles. Are there better options for frequent flyers?"  
+        - Optimized Query: "Category: Travel Benefits | Credit cards with premium travel perks like airport lounge access, hotel discounts, and concierge services. Avoid repeating airline mile features already covered."
+        
         Preferences: "{preferences_text}"
-        User Query: "{user_query}"  
+        User Query: "{user_query}"
+
     """
 
     print("rewriting")
     response = model2.generate_content(prompt)
 
-    print(response.text)
-    return response.text
+    clean_text = re.sub(r"^```json|```$", "", response.text.strip(), flags=re.MULTILINE).strip()
+
+    try:
+        result = json.loads(clean_text)
+        print(result["optimized_query"])
+        print(result.get("excluded_cards", []))
+        return result["optimized_query"], result.get("excluded_cards", [])
+    except Exception as e:
+        print("Error parsing JSON:", e)
+        return clean_text, []
 
 # Function to generate multiple focused subqueries from a user query
 def generate_multi_queries(direct_query, n=3):