1. More structured query format focusing on ICD-10 codes

2. Better parsing of diagnoses from response
3. Weighted confidence scores
4. Specific error handling
5. Focus on clinical differentiation

src/app.py

@@ -144,32 +144,48 @@ def process_speech(new_transcript, history):
         return history
     
     try:
-        # Query the symptom index with user's transcript
-        response = symptom_index.as_query_engine().query(
-            f"""Based on this symptom description, list the top 2-3 most likely ICD-10 codes 
-            and key differentiating questions we should ask: {new_transcript}"""
-        )
+        # First, get potential diagnoses based on symptoms
+        diagnosis_query = f"""
+        Given these symptoms: '{new_transcript}'
         
-        # Parse the response and structure initial findings
-        icd_codes = str(response).strip().split('\n')
-        primary_diagnosis = icd_codes[0] if icd_codes else "Unknown"
+        Identify the most likely ICD-10 diagnoses and key questions to differentiate between them.
+        Focus only on symptoms mentioned and their clinical implications.
+        
+        Format response as:
+        1. Primary suspected diagnosis: [ICD-10 code] - [description]
+        2. Alternative diagnosis: [ICD-10 code] - [description]
+        3. Key differentiating question
+        """
+        
+        response = symptom_index.as_query_engine().query(diagnosis_query)
+        
+        # Parse response into structured format
+        lines = str(response).strip().split('\n')
+        diagnoses = []
+        follow_up = ""
+        
+        for line in lines:
+            if '[' in line and ']' in line:  # Extract ICD-10 codes
+                code = line[line.find('[')+1:line.find(']')]
+                diagnoses.append(code)
+            elif 'Key differentiating question' in line:
+                follow_up = line.split(':')[-1].strip()
         
-        # Structure a response that includes specific follow-up
         formatted_response = {
-            "diagnoses": [primary_diagnosis],
-            "confidences": [0.6],  # Start with moderate confidence
-            "follow_up": f"To differentiate between possible diagnoses, {icd_codes[1] if len(icd_codes) > 1 else 'please describe'}:"
+            "diagnoses": diagnoses[:2],  # Top 2 diagnoses
+            "confidences": [0.7, 0.3] if len(diagnoses) > 1 else [0.7],  # Weighted confidences
+            "follow_up": follow_up if follow_up else "What other symptoms are you experiencing?"
         }
         
-        # Add to conversation history
         history.append({"role": "user", "content": new_transcript})
         history.append({"role": "assistant", "content": json.dumps(formatted_response, indent=2)})
         
     except Exception as e:
+        print(f"Error processing speech: {str(e)}")
         error_response = {
-            "error": str(e),
-            "status": "error",
-            "transcript": new_transcript
+            "diagnoses": ["Error processing symptoms"],
+            "confidences": [0],
+            "follow_up": "Could you please repeat your symptoms?"
         }
         history.append({"role": "user", "content": new_transcript})
         history.append({"role": "assistant", "content": json.dumps(error_response, indent=2)})