Update UI and expanded cards

app.py

@@ -10,9 +10,10 @@ from sentence_transformers import CrossEncoder
 import gradio as gr
 import tempfile
 import re
+import ast
 
 #for adding bank name to the cards in the graph
-eligibility_df = pd.read_csv("cards_eligibility(74 cards).csv")
+eligibility_df = pd.read_csv("cards_eligibility_updated.csv")
 card_to_bank = dict(zip(eligibility_df['Name'], eligibility_df['Bank']))
 
 #neo4j credentials
@@ -37,7 +38,7 @@ def generate_cypher(user_query, query_intent, include_cobranded):
     cypher_prompt = f"""
         You are an expert Neo4j Cypher query generator.
         
-        Given a user’s question, graph schema, and **contextual flags**, generate the correct Cypher query. The query should retrurn only the cards c.
+        Given a user’s question, graph schema, and **contextual flags**, generate the correct Cypher query. The query should return only the cards `c`.
         
         ONLY output the Cypher query. Do NOT explain anything.
         
@@ -49,6 +50,14 @@ def generate_cypher(user_query, query_intent, include_cobranded):
           - (Feature): Properties = name
         - Relationships:
           - (Card)-[:HAS_FEATURE]->(Feature)
+          
+        Feature Inclusion Rules:
+        - Only include relevant features based on user query.
+        - Forex markup fee and foreign transaction fee are the same.
+        - Don’t add “General Cashback” or “General Reward Points” unless explicitly mentioned.
+        - If fuel is mentioned, include both `Fuel Benefits` and `Fuel Surcharge Waiver`.
+        - **ALWAYS** match features using: `f.name IN [...]` — even if there is only **one** feature.
+        
         
         Valid values:
         - card_type: 'FD Card' or 'Regular'
@@ -58,31 +67,62 @@ def generate_cypher(user_query, query_intent, include_cobranded):
         MANDATORY Condition Rules:
         - If FD Card intent is true → include: `c.card_type = 'FD Card'`
         - Else → include: `c.card_type = 'Regular'`
+        - If the query is based on beginners or students or people with no or low credit history then use FD Card.
         - If the query uses words like "premium", "elite", "luxury", "exclusive", "infinia", "black", etc. → include: `AND c.premium = true`
+        - If the query includes low spending, without high spending or budget → include: `(c.premium IS NULL OR c.premium = false)`
         - If include co-branded is false → include: `AND (c.co_branded IS NULL OR c.co_branded = false)`
         - These conditions are **MANDATORY**. If they apply, include them in the `WHERE` clause. Do not skip them.
         
         ---
         
         Available features:
-        "Dining Benefits", "Railway Benefits", "Railway Lounge", "Airport Lounge Access", "Travel Benefits", 
-        "Fuel Surcharge Waiver", "Fuel Benefits", "General Cashback", "General Reward Points", "Movie Benefits", 
-        "Daily Spends (Grocery)", "Utility", "Insurance & Health Benefits", "Forex Markup Fee", "Hotel Benefits", 
-        "Welcome Bonus", "Air Miles", "No Forex Markup Fee", "Flight Discounts"
+          "General Cashback", "Fuel Surcharge Waiver", "Fuel Benefits", "Welcome Bonus",
+          "Airport Lounge Access", "General Reward Points", "Domestic Travel Benefits",
+          "Movie Benefits", "Flight Discounts", "International Travel Benefits",
+          "Hotel Benefits", "Dining Benefits", "Daily Spends (Grocery)", "Railway Benefits",
+          "Travel Benefits", "Railway Lounge", "Insurance", "Utility",
+          "E-commerce Platform Benefits", "Air Miles", "Spa Access Benefits",
+          "Lifestyle & Luxury Perks", "Golf Access & Perks", "Online Shopping Benefits",
+          "UPI Transaction Support", "Health Benefits", "EMI Conversion Options",
+          "No Forex Markup Fee", "Roadside Assistance", "Rupay Network Support"
         
-        Feature Inclusion Rules:
-        - Only include relevant features based on user query.
-        - Don’t add “General Cashback” or “General Reward Points” unless explicitly mentioned.
-        - If fuel is mentioned, include both `Fuel Benefits` and `Fuel Surcharge Waiver`.
+        ---
+        
+        Few-shot Examples:
+        
+        User Query: Show premium cards with airport lounge access  
+        Cypher:
+        MATCH (c:Card)-[:HAS_FEATURE]->(f:Feature)
+        WHERE f.name IN ["Airport Lounge Access"]
+          AND c.card_type = 'Regular'
+          AND c.premium = true
+        RETURN c
+        
+        User Query: I want FD cards with spa access and golf perks  
+        Cypher:
+        MATCH (c:Card)-[:HAS_FEATURE]->(f:Feature)
+        WHERE f.name IN ["Spa Access Benefits", "Golf Access & Perks"]
+          AND c.card_type = 'FD Card'
+        RETURN c
+        
+        User Query: Cards that support UPI but are not co-branded  
+        Cypher:
+        MATCH (c:Card)-[:HAS_FEATURE]->(f:Feature)
+        WHERE f.name IN ["UPI Transaction Support"]
+          AND c.card_type = 'Regular'
+          AND (c.co_branded IS NULL OR c.co_branded = false)
+        RETURN c
         
         ---
         
         {context_note}
         
-        User Query: {user_query}
+        User Query: {user_query}  
         Cypher:
     """
 
+
+
     response = model3.generate_content(cypher_prompt.strip())
     cypher_code = response.text.strip()
 
@@ -95,7 +135,7 @@ def generate_cypher(user_query, query_intent, include_cobranded):
 
     return cypher_code
 
-# print(generate_cypher("Category: Beginner Credit Cards | Best credit cards for first-time users with low fees and easy approval requirements.", True,True))
+# print(generate_cypher("I shop a lot, both online and in stores, and I want a credit card that gives me good cashback on my purchases. Which card would be the best for saving money? I need a budget cxard", True,False))
 
 #generating embeddings (run only once)
 def chunk_text(text, chunk_size=1):
@@ -114,7 +154,7 @@ def get_gemini_embeddings(text_list):
     return np.vstack(embeddings)
 
 # Loading credit card data
-df = pd.read_csv("credit_card_data(74 cards).csv")
+df = pd.read_csv("credit_card_data_updated.csv")
 card_descriptions = dict(zip(df["name"], df["description"]))
 
 # Chunk all card descriptions 
@@ -162,83 +202,7 @@ def eligibility_filter(cards, user_income, user_cibil, user_age,min_joining_fee,
 
     return eligible_cards
 
-#selecting suitable labels for the query
-def classify_user_query(user_query, labels):
-    genai.configure(api_key='AIzaSyAHoi9xbYAThtjXlyF_IKFtruoWYoUCjJQ')  
-    model3 = genai.GenerativeModel('gemini-1.5-flash-latest')
-    print("classifying labels")
-    prompt = f"""
-        You are an assistant that classifies a user's credit card search query into predefined categories.
-        
-        ## Task:
-        Given the user query, select the most relevant labels from the list below. Return only labels that are clearly applicable.
-        
-        ## Label List:
-        {json.dumps(labels, indent=2)}
-        
-        ## Format:
-        Return only JSON like this:
-        {{ "labels": ["Label A", "Label B"] }}
-        
-        ## User Query:
-        "{user_query}"
-    """
-    try:
-        response = model3.generate_content(prompt)
-        text = re.sub(r"^```(?:json)?\s*|\s*```$", "", response.text.strip(), flags=re.IGNORECASE)
-        return json.loads(text).get("labels", [])
-    except Exception as e:
-        print("Error classifying user query:", e)
-        return []
-        
-labels_list = [
-    "Best for International Travel",
-    "Best for Domestic Travel",
-    "Best for Vacation",
-    "Best for International Vacation",
-    "Best for Frequent Flyers",
-    "Low/No Forex Markup Fee",
-    "Best for Air Miles Accumulation",
-    "Best for flight tickets discount",
-    "Airport Lounge Access Priority",
-    "Railway Lounge Access",
-    "Hotel Benefits & Discounts",
-    "Hotel Loyalty Program Integration",
-    "Best for Dining Out / Fine Dining",
-    "Best for Online Food Ordering",
-    "Best for Bars, Pubs & Nightlife",
-    "Dining + Movie Combo Offers",
-    "Best for Lifestyle & Luxury Perks",
-    "Best for Online Shoppers",
-    "Best for Grocery Shopping",
-    "Best for Utility Bill Payments",
-    "Best for Fuel Spends",
-    "Fuel + Grocery Combo",
-    "Best for E-commerce Platforms",
-    "Best for Everyday Spends (Household)",
-    "Best for Cashback",
-    "Flat Cashback Cards",
-    "Best for Reward Points Accumulation",
-    "Accelerated Rewards for Select Categories",
-    "Rotating Category Reward Cards",
-    "Includes Travel Insurance",
-    "Includes Health / Life Insurance",
-    "Purchase Protection & Extended Warranty",
-    "Fraud Liability Coverage",
-    "Concierge & Lifestyle Management Services",
-    "Best Welcome Bonus",
-    "Lifetime Free Card",
-    "Low Annual Fee, High Value",
-    "No Annual Fee",
-    "Best for EMI Conversion",
-    "Best for Air Miles",
-    "Best for Reward Miles",
-    "Best for Movie Ticket Discounts",
-    "Best for Students",
-    "Best for Beginners / First-time Users",
-    "Best for Credit Score Building",
-    "Best for Low Credit Score Applicants"
-]
+
 
 #function for retrieving cards from knowledge graph
 class Neo4jConnectionError(Exception):
@@ -249,19 +213,10 @@ def run_cypher_query(user_query, query, use_eligibility, user_income, user_cibil
     
     try:
         with driver.session() as session:
-            classified_labels = classify_user_query(user_query, labels_list)
-            print("Classified Labels:", classified_labels)
 
             result = session.run(query)
             matched_cards = [record["c"] for record in result]
-
-            if classified_labels:
-                filtered_cards = [
-                    card["name"] for card in matched_cards
-                    if any(label in card.get("labels", []) for label in classified_labels)
-                ]
-            else:
-                filtered_cards = [card["name"] for card in matched_cards]
+            filtered_cards = [card["name"] for card in matched_cards]
 
     except Exception as e:
         raise Neo4jConnectionError("Failed to connect to the Neo4j database.") from e
@@ -380,18 +335,19 @@ def convert_to_direct_query_gradio(user_query,preferences):
         preferences_text = "User selected preferences: " + ", ".join(preferences) + "."
     # Prompt with examples for indirect-to-direct conversion
     prompt = f"""
-        You are an AI assistant that refines user queries to make them optimized for information retrieval** while keeping the original intent.  
+        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.    
+        - Include the preferences also in the final query.  
         - Do NOT introduce new benefits not mentioned by the user.  
-        - If the user has mentioned a vague term try to put it in more direct words covering it's benefits
-        - Even if the query is empty use the preferences to create a query.
-    
+        - 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:  
     
         Example 1  
@@ -408,10 +364,10 @@ def convert_to_direct_query_gradio(user_query,preferences):
     
         Now, optimize the following preferences and user query:  
     
-        
         Preferences: "{preferences_text}"
         User Query: "{user_query}"  
     """
+
     print("rewriting")
     response = model2.generate_content(prompt)
 
@@ -530,7 +486,7 @@ def compare_selected_cards(selected_names, card_lookup):
             return "<b style='color:red;'>Something went wrong while comparing. Please try again.</b>"
         
 # Loading all 55 cards for comparison feature
-df_all_cards = pd.read_csv("credit_card_data(74 cards).csv") 
+df_all_cards = pd.read_csv("credit_card_data_updated.csv") 
 all_card_names = df_all_cards["name"].tolist()
 all_card_lookup = dict(zip(df_all_cards["name"], df_all_cards["description"]))
 
@@ -552,6 +508,7 @@ def find_intent(user_query: str) -> bool:
     response = model2.generate_content(prompt)
     result = response.text.strip().lower()
     return result == "true"
+# print(find_intent("im a beginner to credit cards suggest me suitable cards"))
 
 
 #for queries enquiring about a card
@@ -576,23 +533,23 @@ def handle_query_classification(user_query):
     genai.configure(api_key='AIzaSyDGc7l82E1sa4GarTfGiTXC6dLD7Crk8oo')  
     model1 = genai.GenerativeModel('gemini-1.5-flash-latest')
     prompt = f"""
-        You are a smart financial assistant.
-        
-        ### User's Query:
-        {user_query}
-        
-        ### Task:
-        Classify the user's intent into one of the following categories:
-        1. "retrieve" → If the user is asking for card suggestions, recommendations, or showing cards (e.g., "suggest a card", "need a travel card").
-        2. "specific" → If the user is asking about a particular card by name, the card names need not end with "card" keyword, so also consider that (e.g., "Tell me about HDFC Regalia", "Is SBI Elite good?").
-        3. "no_retrieval" → If the query is generic, chit-chat, or doesn't need credit card data lookup (e.g., "What is interest in credit cards?").
-        
-        Respond in the following JSON format:
-        {{
-          "intent": "retrieve" | "specific" | "no_retrieval",
-          "response": "Only include this if intent is 'no_retrieval'"
-        }}
-    """
+You are a smart financial assistant.
+
+### User's Query:
+{user_query}
+
+### Task:
+Classify the user's intent into one of the following categories:
+1. "retrieve" → If the user is asking for card suggestions, recommendations, or showing cards (e.g., "suggest a card", "need a travel card").
+2. "specific" → If the user is asking about a particular card by name, the card names need not end with "card" keyword, so also consider that (e.g., "Tell me about HDFC Regalia", "Is SBI Elite good?").
+3. "no_retrieval" → If the query is generic, chit-chat, or doesn't need credit card data lookup (e.g., "What is interest in credit cards?").
+
+Respond in the following JSON format:
+{{
+  "intent": "retrieve" | "specific" | "no_retrieval",
+  "response": "Only include this if intent is 'no_retrieval'"
+}}
+"""
 
     raw_response = model1.generate_content(prompt).text.strip()
 
@@ -610,6 +567,8 @@ def handle_query_classification(user_query):
         print("Raw response from LLM:", raw_response)
         raise
 
+# print(handle_query_classification("what is a credit card"))
+
 #passing the card mentioned in the user query
 def find_matching_card(user_query):
     lowered_query = user_query.lower()
@@ -617,7 +576,14 @@ def find_matching_card(user_query):
         if row["name"].lower() in lowered_query:
             return row.to_dict()
     return None
+    
+with open('for_graph_construction_(expanded labels).json') as f:
+    card_feature_data = json.load(f)
 
+card_features_lookup = {
+    card['card_name']: set(card['features'])
+    for card in card_feature_data
+}
 
 #function to pass the retrieved cards and generated response to the UI
 def recommend_cards_gradio(user_query, preferences, income, cibil, age,
@@ -710,18 +676,24 @@ def recommend_cards_gradio(user_query, preferences, income, cibil, age,
                 "No eligible card found"
             )
 
-        card_rows = [
-            [card["name"], card["description"]]
-            for score, card in sorted(
-                zip(cross_encoder.predict([[direct_query, card["description"]] for card in cards]), cards),
-                reverse=True,
-                key=lambda x: x[0]
-            )
-        ]
+        match = re.search(r"f\.name IN (\[.*?\])", cypher_query)
+        query_features = set(ast.literal_eval(match.group(1))) if match else set()
 
+        card_rows = []
+        for score, card in sorted(
+            zip(cross_encoder.predict([[direct_query, card["description"]] for card in cards]), cards),
+            reverse=True,
+            key=lambda x: x[0]
+        ):
+            card_name = card["name"]
+            card_desc = card["description"]
+            matched_features = query_features.intersection(card_features_lookup.get(card_name, set()))
+            feature_str = ", ".join(matched_features) if matched_features else "None"
+            card_rows.append([card_name, feature_str, card_desc])
+
+        
         card_names = [row[0] for row in card_rows]
         card_lookup = {row[0]: row[1] for row in card_rows}
-
         top_card_html = f"""
             <div style="
                 background-color: #fff3e0;
@@ -738,7 +710,7 @@ def recommend_cards_gradio(user_query, preferences, income, cibil, age,
             </div>
         """
 
-        df_cards = pd.DataFrame(card_rows, columns=["Card Name", "Description"])
+        df_cards = pd.DataFrame(card_rows, columns=["Card Name", "Matched Features", "Description"])
         temp_file = tempfile.NamedTemporaryFile(delete=False, suffix=".csv")
         df_cards.to_csv(temp_file.name, index=False)
 
@@ -754,6 +726,7 @@ def recommend_cards_gradio(user_query, preferences, income, cibil, age,
             {},
             "Unexpected error occurred, please try again in a while"
         )
+
     
 #function for the chatbot functionality
 eligibility_lookup = {}
@@ -819,122 +792,120 @@ def chat_with_gemini(user_query, user_message, chat_history, card_lookup):
 
     return chat_history, chat_history
 
-#Interface
+# Interface with Tabs
 with gr.Blocks() as demo:
     gr.Markdown("# Credit Card Recommender")
     gr.Markdown("Get personalized credit card suggestions based on your lifestyle and eligibility.")
-    
-    #Input fields for user query and preferences
-    with gr.Row():
-        user_query = gr.Textbox(
-            label="Enter your query",
-            info="E.g., 'Best cards for international travel' or 'I want cashback cards with lounge access'"
-        )
-        preferences = gr.CheckboxGroup(
-            choices=["Cashback", "Travel Rewards", "Fuel Benefits", "International Lounge access",
-                     "Domestic Lounge access", "Railway benefits", "Dining", "Shopping"],
-            label="Credit card categories:",
-            info="Select the features or benefits you want from your credit card"
-        )
+
+    with gr.Tabs():
         
-    with gr.Accordion("Eligibility filters menu", open=False):
-        with gr.Row():
-            income = gr.Slider(
-                minimum=1, maximum=60, step=1,
-                label="Annual Income (LPA) Minimum requirement is 2.5",
-                info="Helps filter cards based on your income eligibility (in Lakhs Per Annum)"
-            )
-            cibil = gr.Slider(
-                minimum=300, maximum=900, step=10,
-                label="CIBIL Score",
-                info="Most of the cards requires a credit score of 700+"
-            )
-            age = gr.Slider(
-                minimum=18, maximum=75, step=1,
-                label="Age",
-                info="Some cards have minimum and maximum age eligibility"
-            )
-            
-        with gr.Row():
-            min_joining_fee = gr.Number(
-                label="Min Joining Fee (₹)", value=0,
-                info="Minimum one-time fee to get the card"
-            )
-            max_joining_fee = gr.Number(
-                label="Max Joining Fee (₹)", value=150000,
-                info="Maximum one-time fee to get the card"
-            )
-    
-        with gr.Row():
-            min_annual_fee = gr.Number(
-                label="Min Annual Fee (₹)", value=0,
-                info="Minimum yearly fee to be paid"
-            )
-            max_annual_fee = gr.Number(
-                label="Max Annual Fee (₹)", value=150000,
-                info="Maximum yearly fee to be paid"
-            )
+        with gr.Tab("🧠 Get Recommendations"):
+            with gr.Row():
+                user_query = gr.Textbox(
+                    label="Enter your query",
+                    info="E.g., 'Best cards for international travel' or 'I want cashback cards with lounge access'"
+                )
+                preferences = gr.CheckboxGroup(
+                    choices=["Cashback", "Travel Rewards", "Fuel Benefits", "International Lounge access",
+                            "Domestic Lounge access", "Railway benefits", "Dining", "Shopping"],
+                    label="Credit card categories:",
+                    info="Select the features or benefits you want from your credit card"
+                )
 
-    # Toggle for eligibility filter and co-branded cards
-    with gr.Row():
-        use_eligibility = gr.Checkbox(
-            label="Apply Eligibility Filter", value=False,
-            info="Enable this to get recommendations of the cards only for which you are eligible for"
-        )
-        include_cobranded = gr.Checkbox(
-            label="Include Co-branded Cards", value=True,
-            info="Include cards that are co-branded with airlines, retailers, etc."
-        )
+            with gr.Accordion("Eligibility filters menu", open=False):
+                with gr.Row():
+                    income = gr.Slider(
+                        minimum=1, maximum=60, step=1,
+                        label="Annual Income (LPA) Minimum requirement is 2.5",
+                        info="Helps filter cards based on your income eligibility (in Lakhs Per Annum)"
+                    )
+                    cibil = gr.Slider(
+                        minimum=300, maximum=900, step=10,
+                        label="CIBIL Score",
+                        info="Most of the cards requires a credit score of 700+"
+                    )
+                    age = gr.Slider(
+                        minimum=18, maximum=75, step=1,
+                        label="Age",
+                        info="Some cards have minimum and maximum age eligibility"
+                    )
 
-    submit_btn = gr.Button("Recommend Cards", variant='primary')
+                with gr.Row():
+                    min_joining_fee = gr.Number(
+                        label="Min Joining Fee (₹)", value=0,
+                        info="Minimum one-time fee to get the card"
+                    )
+                    max_joining_fee = gr.Number(
+                        label="Max Joining Fee (₹)", value=150000,
+                        info="Maximum one-time fee to get the card"
+                    )
 
-    # Outputs
-    top_card_html = gr.HTML()
-    card_df = gr.Dataframe(headers=["Card Name", "Description"])
-    card_file = gr.File(label="Download Full Recommendations (CSV)")
+                with gr.Row():
+                    min_annual_fee = gr.Number(
+                        label="Min Annual Fee (₹)", value=0,
+                        info="Minimum yearly fee to be paid"
+                    )
+                    max_annual_fee = gr.Number(
+                        label="Max Annual Fee (₹)", value=150000,
+                        info="Maximum yearly fee to be paid"
+                    )
+
+            with gr.Row():
+                use_eligibility = gr.Checkbox(
+                    label="Apply Eligibility Filter", value=False,
+                    info="Enable this to get recommendations of the cards only for which you are eligible for"
+                )
+                include_cobranded = gr.Checkbox(
+                    label="Include Co-branded Cards", value=True,
+                    info="Include cards that are co-branded with airlines, retailers, etc."
+                )
+
+            submit_btn = gr.Button("Recommend Cards", variant='primary')
+
+            top_card_html = gr.HTML()
+            card_df = gr.Dataframe(headers=["Card Name", "Matched Features", "Description"])
+            card_file = gr.File(label="Download Full Recommendations (CSV)")
+
+        with gr.Tab("🔍 Compare Cards"):
+            gr.Markdown("### Compare Recommended Cards")
+            compare_checkboxes = gr.CheckboxGroup(
+                choices=[], label="Select 2 or more cards to compare",
+                info="Pick 2+ cards from the recommended list to see a comparison"
+            )
+            compare_btn = gr.Button("Compare Selected Cards", variant='primary')
+            compare_output = gr.HTML()
+
+            gr.Markdown("### Compare Any Cards from Full List")
+            full_compare_dropdown = gr.Dropdown(
+                choices=all_card_names, multiselect=True, label="Select any 2+ cards",
+                info="Manually compare any cards from the full database"
+            )
+            full_compare_btn = gr.Button("Compare Selected Cards", variant='primary')
+            full_compare_output = gr.HTML()
+
+        with gr.Tab("💬 Ask Follow-up Questions"):
+            gr.Markdown("### Ask any follow-up question ")
+            chatbot = gr.Chatbot(type='messages')
+            user_query_for_chat = gr.Textbox(
+                label="Enter your question",
+                info="Ask follow-ups like 'Which card has better travel insurance?' or 'Which card has less annual fee'"
+            )
+            submit_query_btn = gr.Button("Submit Query", variant='primary')
 
-    # States for recommendations and chatbot
     card_names_state = gr.State()
     card_lookup_state = gr.State()
     chat_history = gr.State([]) 
-    query=gr.State([])
-    
-    # Recommended cards comparison 
-    gr.Markdown("### Compare Recommended Cards")
-    compare_checkboxes = gr.CheckboxGroup(
-        choices=[], label="Select 2 or more cards to compare",
-        info="Pick 2+ cards from the recommended list to see a comparison"
-    )
-    compare_btn = gr.Button("Compare Selected Cards", variant='primary')
-    compare_output = gr.HTML()
-
-    # Full card list comparison
-    gr.Markdown("### Compare Any Cards from Full List")
-    full_compare_dropdown = gr.Dropdown(
-        choices=all_card_names, multiselect=True, label="Select any 2+ cards",
-        info="Manually compare any cards from the full database"
-    )
-    full_compare_btn = gr.Button("Compare Selected Cards", variant='primary')
-    full_compare_output = gr.HTML()
-
-    # Chatbot UI for follow-up questions
-    gr.Markdown("### Ask any follow-up question ")
-    chatbot = gr.Chatbot(type='messages')
+    query = gr.State([])
 
-    # Function for passing inputs and generating the results
     def wrapped_recommend_cards(user_query, preferences, income, cibil, age, min_joining_fee, max_joining_fee,
                         min_annual_fee, max_annual_fee, use_eligibility,include_cobranded):
         top_html, df, file, card_names, card_lookup, direct_query = recommend_cards_gradio(
             user_query, preferences, income, cibil, age, min_joining_fee, max_joining_fee,
             min_annual_fee, max_annual_fee, use_eligibility,include_cobranded
-            )
-        
+        )
         df_label = f"Found {len(card_names)} cards"
-        
         return top_html, gr.update(value=df, label=df_label), file, card_names, card_lookup, gr.update(choices=card_names, value=[]), direct_query
 
-
-    # Button actions and UI interactions
     submit_btn.click(
         fn=wrapped_recommend_cards,
         inputs=[user_query, preferences, income, cibil, age, min_joining_fee, max_joining_fee,
@@ -954,17 +925,11 @@ with gr.Blocks() as demo:
         outputs=full_compare_output
     )
 
-    # Chatbot function for handling follow-up queries
-    user_query_for_chat = gr.Textbox(
-        label="Enter your question",
-        info="Ask follow-ups like 'Which card has better travel insurance?' or 'Which card has less annual fee'"
-    )
-    submit_query_btn = gr.Button("Submit Query", variant='primary')
-
     submit_query_btn.click(
         fn=chat_with_gemini, 
         inputs=[query,user_query_for_chat, chat_history, card_lookup_state],
         outputs=[chatbot, chat_history]
     )
 
+
 demo.launch(share=True)