Update intent_classification/retrieval_classification.py

intent_classification/retrieval_classification.py

@@ -1,80 +1,80 @@
-import google.generativeai as genai
-import pandas as pd
-import os
-import json
-from data import df_all_cards
-
-#handling intent classification for retrieval
-def handle_query_classification(user_query):
-    genai.configure(api_key=os.environ.get("api_key_1"))  
-    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") OR if they mention their lifestyle, income, spending, or needs (e.g., travel, shopping, fuel, rewards, luxury).
-        2. "specific" → If the user is asking about a particular credit card by name (even if the word "card" is not used). Examples: "Tell me about HDFC Regalia", "Is SBI Elite good?".
-        3. "no_retrieval" → ONLY if the query is generic (e.g., “What is credit score?”), casual chit-chat (e.g., “Hi”), or doesn’t mention any lifestyle, financial needs, or specific card names.
-
-        
-        Respond ONLY in the following JSON format:
-        If intent is "no_retrieval", you MUST include a helpful 'response' field.
-        If intent is "retrieve" or "specific", do NOT include any response or explanation.
-        
-        Respond in this exact format:
-        {{
-          "intent": "retrieve" | "specific" | "no_retrieval",
-          "response": "Only include this if intent is 'no_retrieval'"
-        }}
-        
-        """
-
-    raw_response = model1.generate_content(prompt).text.strip()
-
-    # Clean any markdown formatting if present
-    if raw_response.startswith("```"):
-        raw_response = raw_response.strip("`").strip()
-        if raw_response.startswith("json"):
-            raw_response = raw_response[len("json"):].strip()
-
-    try:
-        parsed = json.loads(raw_response)
-        return parsed
-    except Exception as e:
-        print("JSON parsing error:", e)
-        print("Raw response from LLM:", raw_response)
-        raise
-# result = handle_query_classification("Want to optimize my spending – travel often, premium hotels, and online shopping.")
-# if result["intent"] == "no_retrieval":
-#     print(result['response'])
-
-#passing the card mentioned in the user query
-def find_matching_card(user_query):
-    lowered_query = user_query.lower()
-    for _, row in df_all_cards.iterrows():
-        if row["name"].lower() in lowered_query:
-            return row.to_dict()
-    return None
-
-
-#for queries enquiring about a card
-def generate_card_response_with_context(user_query, card_info):
-    genai.configure(api_key=os.environ.get("api_key_1"))  
-    model1 = genai.GenerativeModel('gemini-1.5-flash-latest')
-    prompt = f"""
-You are a helpful financial assistant. A user has asked about a specific credit card.
-
-Card Name: {card_info.get('name')}
-Description: {card_info.get('description')}
-
-User's Question: {user_query}
-
-Please provide a concise, relevant answer using the above card context.
-"""
-    response = model1.generate_content(prompt)
-    return response.text.strip()
-
+import google.generativeai as genai
+import pandas as pd
+import os
+import json
+from data import df_all_cards
+
+#handling intent classification for retrieval
+def handle_query_classification(user_query):
+    genai.configure(api_key=os.environ.get("api_key_1"))  
+    model1 = genai.GenerativeModel('gemini-2.0-flash')
+    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") OR if they mention their lifestyle, income, spending, or needs (e.g., travel, shopping, fuel, rewards, luxury).
+        2. "specific" → If the user is asking about a particular credit card by name (even if the word "card" is not used). Examples: "Tell me about HDFC Regalia", "Is SBI Elite good?".
+        3. "no_retrieval" → ONLY if the query is generic (e.g., “What is credit score?”), casual chit-chat (e.g., “Hi”), or doesn’t mention any lifestyle, financial needs, or specific card names.
+
+        
+        Respond ONLY in the following JSON format:
+        If intent is "no_retrieval", you MUST include a helpful 'response' field.
+        If intent is "retrieve" or "specific", do NOT include any response or explanation.
+        
+        Respond in this exact format:
+        {{
+          "intent": "retrieve" | "specific" | "no_retrieval",
+          "response": "Only include this if intent is 'no_retrieval'"
+        }}
+        
+        """
+
+    raw_response = model1.generate_content(prompt).text.strip()
+
+    # Clean any markdown formatting if present
+    if raw_response.startswith("```"):
+        raw_response = raw_response.strip("`").strip()
+        if raw_response.startswith("json"):
+            raw_response = raw_response[len("json"):].strip()
+
+    try:
+        parsed = json.loads(raw_response)
+        return parsed
+    except Exception as e:
+        print("JSON parsing error:", e)
+        print("Raw response from LLM:", raw_response)
+        raise
+# result = handle_query_classification("Want to optimize my spending – travel often, premium hotels, and online shopping.")
+# if result["intent"] == "no_retrieval":
+#     print(result['response'])
+
+#passing the card mentioned in the user query
+def find_matching_card(user_query):
+    lowered_query = user_query.lower()
+    for _, row in df_all_cards.iterrows():
+        if row["name"].lower() in lowered_query:
+            return row.to_dict()
+    return None
+
+
+#for queries enquiring about a card
+def generate_card_response_with_context(user_query, card_info):
+    genai.configure(api_key=os.environ.get("api_key_1"))  
+    model1 = genai.GenerativeModel('gemini-2.0-flash')
+    prompt = f"""
+You are a helpful financial assistant. A user has asked about a specific credit card.
+
+Card Name: {card_info.get('name')}
+Description: {card_info.get('description')}
+
+User's Question: {user_query}
+
+Please provide a concise, relevant answer using the above card context.
+"""
+    response = model1.generate_content(prompt)
+    return response.text.strip()
+