Update main.py

main.py

@@ -171,20 +171,23 @@ class IrisReportEngine:
 
     def synthesize_fallback_response(self, briefing: dict, user_question: str) -> str:
         fallback_prompt = f"""
-        You are Iris, an expert business data analyst. Your primary role is to provide intelligent insights and help the user with their business.
-        You were unable to process a complex user query. Do not mention the error. Instead, gracefully pivot by presenting a "Business Intelligence Briefing".
-        Structure your response with clear markdown headings for each section of the briefing data.
-        Crucially, interpret the data in the "Performance Snapshot" - highlight the percentage changes as indicators of trends (e.g., "Revenue is up by 15.2%...").
-
-        using the business data also provide insight and suggest improvements and ideas where necessary.
+        You are Iris, an expert business data analyst. Answer the user's question using the comprehensive business data below.
+        
+        If their question is specific (like "sales yesterday", "top product", etc.), directly answer it using the data.
+        If you cannot find the specific information requested, provide a helpful business intelligence briefing instead.
+        
+        Structure your response with clear markdown headings and focus on actionable insights.
+        Always interpret percentage changes as business trends and provide context.
         
-        User's Original Question: "{user_question}"
-        Business Intelligence Briefing Data: {json.dumps(briefing, indent=2, ensure_ascii=False)}
+        User's Question: "{user_question}"
+        Business Data: {json.dumps(briefing, indent=2, ensure_ascii=False)}
         """
         response = self.llm.invoke(fallback_prompt)
         return response.content if hasattr(response, 'content') else str(response)
 
-# --- REFACTORED /chat Endpoint with Correct Tiered Logic ---
+# REMOVED: No error detection function needed - Trust PandasAI completely, catch ALL exceptions silently
+
+# --- REFACTORED /chat Endpoint with Enhanced Error Detection ---
 @app.route("/chat", methods=["POST"])
 @cross_origin()
 def bot():
@@ -202,50 +205,79 @@ def bot():
         transactions = response.json().get("transactions")
         if not transactions: return jsonify({"answer": "No transaction data was found for this profile."})
 
-        # --- TIER 1 (DEFAULT): PANDASAI FIRST ---
+        # --- TIER 1 (DEFAULT): PANDASAI FIRST - WITH COMPREHENSIVE RESPONSE VALIDATION ---
         try:
-            logger.info("Attempting to answer with Tier 1 (PandasAI)...")
+            logger.info("Attempting to answer with Tier 1 (PandasAI) - Full Trust Mode...")
             df = pd.DataFrame(transactions)
             
-            # START: PRESERVED PANDASAI IMPLEMENTATION
+            # FULL TRUST PANDASAI CONFIGURATION
             pandas_agent = SmartDataframe(df, config={
-                "llm": llm, "response_parser": FlaskResponse,
+                "llm": llm, 
+                "response_parser": FlaskResponse,
                 "custom_whitelisted_dependencies": [
-                    "os", "io", "sys", "chr", "glob",
-                    "b64decoder", "collections", "geopy",
-                    "geopandas", "wordcloud", "builtins"
+                    "os", "io", "sys", "chr", "glob", "b64decoder", "collections", 
+                    "geopy", "geopandas", "wordcloud", "builtins", "datetime", 
+                    "timedelta", "date", "pandas", "numpy", "math", "statistics",
+                    "matplotlib", "seaborn", "plotly", "json", "re", "warnings"
                 ],
-                "security": "none", "save_charts_path": user_defined_path, 
-                "save_charts": False, "enable_cache": False, "conversational":True
+                "security": "none",
+                "save_charts_path": user_defined_path, 
+                "save_charts": False, 
+                "enable_cache": False, 
+                "conversational": True,
+                "enable_logging": False
             })
+            
             answer = pandas_agent.chat(user_question)
             
-            # ROBUSTNESS CHECK: Actively inspect the answer for soft failures.
-            is_failure = False
-            if answer is None:
-                is_failure = True
-                logger.warning("PandasAI returned None. Triggering fallback.")
-            if isinstance(answer, str):
-                fail_strings = ["i am sorry", "i cannot answer", "an error occurred", "unable to answer"]
-                if any(s in answer.lower() for s in fail_strings):
-                    is_failure = True
-                    logger.warning(f"PandasAI returned a failure string: '{answer}'. Triggering fallback.")
+            # COMPREHENSIVE RESPONSE VALIDATION - Check if PandasAI actually succeeded
+            # PandasAI doesn't raise exceptions, it returns responses that may contain errors
+            is_valid_response = True
             
-            if not is_failure:
+            # Check 1: Answer exists and is not empty
+            if answer is None or (isinstance(answer, str) and not answer.strip()):
+                is_valid_response = False
+            
+            # Check 2: Answer doesn't contain error indicators (PandasAI returns these as strings)
+            elif isinstance(answer, str):
+                error_patterns = [
+                    'keyerror', 'traceback', 'exception', 'error occurred', 
+                    'failed', 'unable to', 'cannot', '__import__', 'importerror',
+                    'modulenotfounderror', 'nameerror', 'syntaxerror',
+                    'pipeline failed', 'execution failed'
+                ]
+                answer_lower = answer.lower()
+                if any(pattern in answer_lower for pattern in error_patterns):
+                    is_valid_response = False
+                    
+                # Also check for stack traces or error messages that slip through
+                if 'file "<string>"' in answer_lower or 'line ' in answer_lower and 'error' in answer_lower:
+                    is_valid_response = False
+            
+            # Check 3: For specific error objects that might be returned
+            elif hasattr(answer, '__class__') and 'error' in str(type(answer)).lower():
+                is_valid_response = False
+            
+            if is_valid_response:
                 logger.info("Successfully answered with Tier 1 (PandasAI).")
                 formatted_answer = str(answer)
-                if isinstance(answer, pd.DataFrame): formatted_answer = answer.to_html()
+                if isinstance(answer, pd.DataFrame): 
+                    formatted_answer = answer.to_html()
                 elif isinstance(answer, plt.Figure):
                     buf = io.BytesIO()
                     answer.savefig(buf, format="png")
                     formatted_answer = f"data:image/png;base64,{base64.b64encode(buf.getvalue()).decode('utf-8')}"
                 return jsonify({"answer": formatted_answer})
-
+            else:
+                logger.info("PandasAI response contains error indicators, using analyst layer")
+        
         except Exception as e:
-            logger.warning(f"Tier 1 (PandasAI) failed with exception: '{e}'. Proceeding to Tier 2 Fallback.")
+            # This catches any actual exceptions that might escape PandasAI
+            logger.info(f"PandasAI raised exception, seamlessly switching to analyst layer: {type(e).__name__}")
+            pass
         
-        # --- TIER 2 (GRACEFUL FALLBACK): COMPREHENSIVE KPI ANALYST ---
-        logger.info("Executing Tier 2 Fallback: IrisReportEngine.")
+        # --- TIER 2 (SEAMLESS FALLBACK): COMPREHENSIVE KPI ANALYST ---
+        logger.info("Seamlessly providing intelligence via IrisReportEngine analyst layer.")
         engine = IrisReportEngine(transactions_data=transactions, llm_instance=llm)
         briefing = engine.get_business_intelligence_briefing()
         fallback_answer = engine.synthesize_fallback_response(briefing, user_question)