fix: emergency issue

app/graph.py

@@ -112,9 +112,15 @@ def agent_node(state: IntakeState) -> dict:
     current_json = state.get("clinical_state") or CombinedOutput().model_dump_json()
     transcript = format_transcript(msgs)
 
+    import time
+    t_agent = time.time()
+    print(f"[{time.time():.3f}] [Graph Node] Requesting LLM inference...")
+
     llm = get_llm()
     result: CombinedOutput = llm.combined_call(transcript, current_json)
 
+    print(f"[{time.time():.3f}] [Graph Node] LLM returned. Preparing node dictionaries...")
+
     if result.emergency:
         return {
             "messages": [{"role": "assistant", "content": (

app/llm.py

@@ -14,7 +14,7 @@ CRITICAL RULES:
 - Do NOT diagnose or give medical advice.
 - Do NOT ask more than one question.
 - If all fields are complete, set reply to "Thank you — I have everything I need."
-- Emergency override: if patient mentions "crushing chest pain", "can't breathe", "suicide", or similar life-threatening phrases, set emergency=true.
+- EMERGENCY RULE: Set emergency=true ONLY IF the patient explicitly describes an acute life-threatening event right now (e.g., EXACT words "crushing chest pain", "I can't breathe right now", "I want to kill myself"). Do NOT set emergency=true for generic "chest pain", "headache", or past symptoms. Default is always false.
 
 OUTPUT FORMAT (strictly follow this, no extra text):
 {

app/main.py

@@ -82,26 +82,38 @@ async def health():
 
 @app.post("/chat", response_model=ChatResponse)
 async def chat(request: ChatRequest):
+    import time
+    t0 = time.time()
+    print(f"\n[{t0:.3f}] [API] -> POST /chat received for {request.session_id}")
     config = {"configurable": {"thread_id": request.session_id}}
     
     # Get current checkpoint state
     snapshot = graph.get_state(config)
+    print(f"[{time.time():.3f}] [API] Read existing state snapshot.")
     
     # Check if graph is interrupted and waiting for input
+    t_start_graph = time.time()
     if snapshot.next:
+        print(f"[{time.time():.3f}] [API] Resuming graph from interrupt...")
         # First update state with the user message
         graph.update_state(config, {"messages": [{"role": "user", "content": request.message}]})
         # Then resume execution
         result = graph.invoke(None, config=config)
     else:
+        print(f"[{time.time():.3f}] [API] Starting new graph invoke...")
         # New conversation - start fresh
         input_state = {"messages": [{"role": "user", "content": request.message}]}
         result = graph.invoke(input_state, config=config)
+    print(f"[{time.time():.3f}] [API] <- Graph invoke returned in {time.time() - t_start_graph:.2f}s")
     
+    t_final = time.time()
     current_node = get_current_node(request.session_id)
     reply = get_last_reply(request.session_id)
     brief_dict = get_brief(request.session_id)
     
+    total_t = time.time() - t0
+    print(f"[{time.time():.3f}] [API] Chat completed in {total_t:.2f}s total. Reply length: {len(reply)}")
+    
     return ChatResponse(reply=reply, state=current_node, brief=brief_dict)