Update app.py

app.py

@@ -28,17 +28,15 @@ def analyze_attrition_with_llm(df_dict, hr_query):
         return "❌ Error: No processed data. Upload a CSV first."
     
     df = df_dict["df"]
-    employees_data = [
-        {"employee_name": row["Employee"], "sentiment": row["Sentiment"], "feedback": row["Feedback"]}
-        for _, row in df.iterrows()
-    ]
-    
-    prompt = f"HR asked: '{hr_query}'. Here is the employee sentiment data:\n{json.dumps(employees_data, indent=2)}\n"
-    prompt += "If an employee is not found, return 'No records found for this employee.' If the question is irrelevant, refuse to answer politely."
+    employees_data = {row["Employee"].strip(): row["Sentiment"] for _, row in df.iterrows()}
     
+    # Use GPT-4-turbo to analyze the HR query
     response = client.chat.completions.create(
         model="gpt-4-turbo",
-        messages=[{"role": "user", "content": prompt}],
+        messages=[
+            {"role": "system", "content": "You are an HR assistant. Only respond to queries about employee attrition risk based on sentiment. If the query is irrelevant, reply with an apology."},
+            {"role": "user", "content": hr_query}
+        ],
         functions=[
             {
                 "name": "predict_attrition_risk",
@@ -46,8 +44,8 @@ def analyze_attrition_with_llm(df_dict, hr_query):
                 "parameters": {
                     "type": "object",
                     "properties": {
-                        "employee_name": {"type": "string"},
-                        "sentiment": {"type": "string"}
+                        "employee_name": {"type": "string", "description": "Employee's name"},
+                        "sentiment": {"type": "string", "description": "Extracted sentiment"}
                     },
                     "required": ["employee_name", "sentiment"]
                 }
@@ -57,24 +55,20 @@ def analyze_attrition_with_llm(df_dict, hr_query):
     )
     
     message = response.choices[0].message
-    
-    if hasattr(message, "function_call") and message.function_call:
+    if hasattr(message, "function_call") and message.function_call is not None:
         try:
-            function_calls = json.loads(message.function_call.arguments)
-            results = []
-            for call in function_calls if isinstance(function_calls, list) else [function_calls]:
-                employee_name = call.get("employee_name")
-                sentiment = call.get("sentiment")
-                
-                if not any(emp["employee_name"] == employee_name for emp in employees_data):
-                    return "No records found for this employee."
-                
-                results.append(predict_attrition_risk(employee_name, sentiment))
-            return "\n".join(results)
-        except Exception:
-            return "No records found for this employee."
+            function_call = json.loads(message.function_call.arguments)
+            employee_name = function_call.get("employee_name")
+            sentiment = employees_data.get(employee_name)
+            
+            if sentiment:
+                return predict_attrition_risk(employee_name, sentiment)
+            else:
+                return f"{employee_name}: No records found for this employee."
+        except Exception as e:
+            return f"❌ Error processing LLM function call: {str(e)}"
     
-    return "I'm sorry, but I can only answer HR-related questions about employee sentiment and attrition risk." 
+    return "🤖 I'm sorry, but I can only answer queries related to employee attrition risk."
 
 with gr.Blocks() as demo:
     gr.Markdown("<h1>AI-Driven Employee Attrition Risk Analysis</h1>")