Update api/server.py

api/server.py

@@ -42,7 +42,7 @@ WEB_INDEX = os.path.join(WEB_DIST, "index.html")
 WEB_ASSETS = os.path.join(WEB_DIST, "assets")
 
 LS_DATASET_NAME = os.getenv("LS_DATASET_NAME", "clare_user_events").strip()
-LS_PROJECT = os.getenv("LANGSMITH_PROJECT", os.getenv("LANGCHAIN_PROJECT", "")).strip()  # optional
+LS_PROJECT = os.getenv("LANGSMITH_PROJECT", os.getenv("LANGCHAIN_PROJECT", "")).strip()
 
 EXPERIMENT_ID = os.getenv("CLARE_EXPERIMENT_ID", "RESP_AI_W10").strip()
 
@@ -55,13 +55,10 @@ WARMUP_DONE = False
 WARMUP_ERROR: Optional[str] = None
 WARMUP_STARTED = False
 
-# warmup knobs
 CLARE_ENABLE_WARMUP = os.getenv("CLARE_ENABLE_WARMUP", "1").strip() == "1"
 CLARE_WARMUP_BLOCK_READY = os.getenv("CLARE_WARMUP_BLOCK_READY", "0").strip() == "1"
 
-# langsmith knobs (important for latency)
 CLARE_ENABLE_LANGSMITH_LOG = os.getenv("CLARE_ENABLE_LANGSMITH_LOG", "0").strip() == "1"
-# If true, logging is done in background thread to avoid blocking /api/chat
 CLARE_LANGSMITH_ASYNC = os.getenv("CLARE_LANGSMITH_ASYNC", "1").strip() == "1"
 
 # ----------------------------
@@ -113,7 +110,6 @@ def _preload_module10_chunks() -> List[Dict[str, Any]]:
     return []
 
 
-# Preload at import time (fast path for requests)
 MODULE10_CHUNKS_CACHE = _preload_module10_chunks()
 
 
@@ -133,28 +129,18 @@ def _get_session(user_id: str) -> Dict[str, Any]:
 
 
 # ----------------------------
-# Warmup (runs once, background)
+# Warmup
 # ----------------------------
 def _do_warmup_once():
-    """
-    Warm OpenAI connection + touch module10 chunks cache.
-    Best-effort; should never crash the app.
-    """
     global WARMUP_DONE, WARMUP_ERROR, WARMUP_STARTED
     if WARMUP_STARTED:
         return
     WARMUP_STARTED = True
 
     try:
-        # Warm OpenAI network / TLS / keep-alive
         from api.config import client
-
-        # cheapest call: models.list() (no token usage)
         client.models.list()
-
-        # Touch module10 cache (already loaded at import; this is just a safety)
         _ = MODULE10_CHUNKS_CACHE
-
         WARMUP_DONE = True
         WARMUP_ERROR = None
     except Exception as e:
@@ -174,7 +160,7 @@ def _on_startup():
 
 
 # ----------------------------
-# LangSmith helpers (optional; default OFF)
+# LangSmith helpers
 # ----------------------------
 _ls_client = None
 if (Client is not None) and CLARE_ENABLE_LANGSMITH_LOG:
@@ -186,10 +172,6 @@ if (Client is not None) and CLARE_ENABLE_LANGSMITH_LOG:
 
 
 def _log_event_to_langsmith(data: Dict[str, Any]):
-    """
-    Create an Example in LangSmith Dataset.
-    Best-effort and non-blocking by default (async thread).
-    """
     if _ls_client is None:
         return
 
@@ -201,6 +183,8 @@ def _log_event_to_langsmith(data: Dict[str, Any]):
                 "student_name": data.get("student_name", ""),
             }
             outputs = {"answer": data.get("answer", "")}
+
+            # keep metadata clean and JSON-serializable
             metadata = {k: v for k, v in data.items() if k not in ("question", "answer")}
 
             if LS_PROJECT:
@@ -222,11 +206,10 @@ def _log_event_to_langsmith(data: Dict[str, Any]):
 
 
 # ----------------------------
-# Health endpoints (pure lightweight)
+# Health endpoints
 # ----------------------------
 @app.get("/health")
 def health():
-    # do not touch LLM/RAG/disk heavy work here
     return {
         "ok": True,
         "uptime_s": round(time.time() - APP_START_TS, 3),
@@ -243,13 +226,10 @@ def health():
 
 @app.get("/ready")
 def ready():
-    # readiness probe: optionally block until warmup completes
     if not CLARE_ENABLE_WARMUP or not CLARE_WARMUP_BLOCK_READY:
         return {"ready": True}
-
     if WARMUP_DONE:
         return {"ready": True}
-
     return JSONResponse({"ready": False, "error": WARMUP_ERROR}, status_code=503)
 
 
@@ -281,6 +261,10 @@ class SummaryReq(BaseModel):
 
 
 class FeedbackReq(BaseModel):
+    # IMPORTANT: allow extra fields so FE can evolve without breaking backend
+    class Config:
+        extra = "ignore"
+
     user_id: str
     rating: str  # "helpful" | "not_helpful"
     assistant_message_id: Optional[str] = None
@@ -288,11 +272,12 @@ class FeedbackReq(BaseModel):
     assistant_text: str
     user_text: Optional[str] = ""
 
-    tags: Optional[List[str]] = []  # ✅ NEW: user selected chips/tags
-
     comment: Optional[str] = ""
 
+    # optional structured fields
+    tags: Optional[List[str]] = []
     refs: Optional[List[str]] = []
+
     learning_mode: Optional[str] = None
     doc_type: Optional[str] = None
     timestamp_ms: Optional[int] = None
@@ -332,33 +317,24 @@ def chat(req: ChatReq):
             "latency_ms": 0.0,
         }
 
-    # ----------------------------
-    # Latency breakdown marks (ms)
-    # ----------------------------
     t0 = time.time()
     marks_ms: Dict[str, float] = {"start": 0.0}
 
-    # language detect
     resolved_lang = detect_language(msg, req.language_preference)
     marks_ms["language_detect_done"] = (time.time() - t0) * 1000.0
 
-    # weakness update
     sess["weaknesses"] = update_weaknesses_from_message(msg, sess["weaknesses"])
     marks_ms["weakness_update_done"] = (time.time() - t0) * 1000.0
 
-    # cognitive update
     sess["cognitive_state"] = update_cognitive_state_from_message(msg, sess["cognitive_state"])
     marks_ms["cognitive_update_done"] = (time.time() - t0) * 1000.0
 
-    # rag retrieve (optional micro-gate for very short messages)
     if len(msg) < 20 and ("?" not in msg):
         rag_context_text, rag_used_chunks = "", []
     else:
         rag_context_text, rag_used_chunks = retrieve_relevant_chunks(msg, sess["rag_chunks"])
-
     marks_ms["rag_retrieve_done"] = (time.time() - t0) * 1000.0
 
-    # llm
     try:
         answer, new_history = chat_with_clare(
             message=msg,
@@ -379,7 +355,6 @@ def chat(req: ChatReq):
     marks_ms["llm_done"] = (time.time() - t0) * 1000.0
     total_ms = marks_ms["llm_done"]
 
-    # segments (delta)
     ordered = [
         "start",
         "language_detect_done",
@@ -403,12 +378,10 @@ def chat(req: ChatReq):
         for c in (rag_used_chunks or [])
     ]
 
-    # extra metadata fields
     rag_context_chars = len(rag_context_text or "")
     rag_used_chunks_count = len(rag_used_chunks or [])
     history_len = len(sess["history"])
 
-    # ✅ log chat_turn to LangSmith (optional; async by default)
     _log_event_to_langsmith(
         {
             "experiment_id": EXPERIMENT_ID,
@@ -514,6 +487,14 @@ def api_feedback(req: FeedbackReq):
     if rating not in ("helpful", "not_helpful"):
         return JSONResponse({"ok": False, "error": "Invalid rating"}, status_code=400)
 
+    # normalize fields
+    assistant_text = (req.assistant_text or "").strip()
+    user_text = (req.user_text or "").strip()
+    comment = (req.comment or "").strip()
+    refs = req.refs or []
+    tags = req.tags or []
+    timestamp_ms = int(req.timestamp_ms or int(time.time() * 1000))
+
     _log_event_to_langsmith(
         {
             "experiment_id": EXPERIMENT_ID,
@@ -521,16 +502,20 @@ def api_feedback(req: FeedbackReq):
             "student_name": student_name,
             "event_type": "feedback",
             "timestamp": time.time(),
+            "timestamp_ms": timestamp_ms,
             "rating": rating,
             "assistant_message_id": req.assistant_message_id,
-            "question": (req.user_text or "").strip(),
-            "answer": (req.assistant_text or "").strip(),
-            "tags": req.tags or [],  # ✅ NEW
-            "comment": (req.comment or "").strip(),
-            "refs": req.refs or [],
+
+            # Keep the Example readable:
+            "question": user_text,            # what user asked (optional)
+            "answer": assistant_text,         # the assistant response being rated
+
+            # metadata
+            "comment": comment,
+            "tags": tags,
+            "refs": refs,
             "learning_mode": req.learning_mode,
             "doc_type": req.doc_type,
-            "timestamp_ms": req.timestamp_ms,
         }
     )