json is not needed at all for this

app.py

@@ -5,7 +5,6 @@ from utils.voice_input_utils import update_live_transcription, format_response_f
 from services.embeddings import configure_embeddings
 from services.indexing import create_symptom_index
 import torchaudio.transforms as T
-import json
 import re
 import logging, sys
 
@@ -95,11 +94,13 @@ def on_submit(symptoms_text, history):
         context_list.append(f"{md.get('code','')}: {md.get('description','')}")
     context_text = "\n".join(context_list)
     prompt = (
-        f"{SYSTEM_PROMPT}\n\n"
-        f"User symptoms: '{cleaned}'\n\n"
-        f"Relevant ICD-10 context:\n{context_text}\n\n"
-        "Respond with valid JSON."
+        f"{SYSTEM_PROMPT}\n\n",
+        f"User symptoms: '{cleaned}'\n\n",
+        f"Relevant ICD-10 context:\n{context_text}\n\n",
+        "Respond with your top 3 ICD-10 codes and their confidence scores.\n\n",
+        "Think step by step and explain your reasoning.\n",
     )
+
     msg = "✏️ Prompt built"
     log.append(msg)
     print(msg, flush=True)
@@ -114,12 +115,6 @@ def on_submit(symptoms_text, history):
     print(msg, flush=True)
     yield history, None, "\n".join(log)
 
-    # Parse JSON
-    cleaned_raw = re.sub(r",\s*([}\]])", r"\1", raw)
-    log.append(msg)
-    print(msg, flush=True)
-    yield history, raw, "\n".join(log)
-
     # Final assistant message
     assistant_msg = format_response_for_user(raw)
     history = history + [{"role": "assistant", "content": assistant_msg}]