Update api/server.py

api/server.py

@@ -1,8 +1,7 @@
 # api/server.py
 import os
 import time
-import json
-from typing import Dict, List, Optional
+from typing import Dict, List, Optional, Any, Tuple
 
 from fastapi import FastAPI, UploadFile, File, Form, Request
 from fastapi.responses import FileResponse, JSONResponse
@@ -23,12 +22,15 @@ from api.clare_core import (
     summarize_conversation,
 )
 
+# ✅ LangSmith
 try:
     from langsmith import Client
 except Exception:
     Client = None
 
-
+# ----------------------------
+# Paths / Constants
+# ----------------------------
 API_DIR = os.path.dirname(__file__)
 
 MODULE10_PATH = os.path.join(API_DIR, "module10_responsible_ai.pdf")
@@ -39,8 +41,13 @@ 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()
+LS_PROJECT = os.getenv("LANGSMITH_PROJECT", os.getenv("LANGCHAIN_PROJECT", "")).strip()  # optional
+
+EXPERIMENT_ID = os.getenv("CLARE_EXPERIMENT_ID", "RESP_AI_W10").strip()
 
+# ----------------------------
+# App
+# ----------------------------
 app = FastAPI(title="Clare API")
 
 app.add_middleware(
@@ -51,6 +58,9 @@ app.add_middleware(
     allow_headers=["*"],
 )
 
+# ----------------------------
+# Static hosting (Vite build)
+# ----------------------------
 if os.path.isdir(WEB_ASSETS):
     app.mount("/assets", StaticFiles(directory=WEB_ASSETS), name="assets")
 
@@ -68,10 +78,13 @@ def index():
     )
 
 
-SESSIONS: Dict[str, Dict] = {}
+# ----------------------------
+# In-memory session store (MVP)
+# ----------------------------
+SESSIONS: Dict[str, Dict[str, Any]] = {}
 
 
-def _preload_module10_chunks():
+def _preload_module10_chunks() -> List[Dict[str, Any]]:
     if os.path.exists(MODULE10_PATH):
         try:
             return build_rag_chunks_from_file(MODULE10_PATH, MODULE10_DOC_TYPE) or []
@@ -84,12 +97,12 @@ def _preload_module10_chunks():
 MODULE10_CHUNKS_CACHE = _preload_module10_chunks()
 
 
-def _get_session(user_id: str) -> Dict:
+def _get_session(user_id: str) -> Dict[str, Any]:
     if user_id not in SESSIONS:
         SESSIONS[user_id] = {
             "user_id": user_id,
             "name": "",
-            "history": [],
+            "history": [],  # List[Tuple[str, str]]
             "weaknesses": [],
             "cognitive_state": {"confusion": 0, "mastery": 0},
             "course_outline": DEFAULT_COURSE_TOPICS,
@@ -99,6 +112,9 @@ def _get_session(user_id: str) -> Dict:
     return SESSIONS[user_id]
 
 
+# ----------------------------
+# LangSmith helpers
+# ----------------------------
 _ls_client = None
 if Client is not None:
     try:
@@ -108,10 +124,12 @@ if Client is not None:
         _ls_client = None
 
 
-def _log_event_to_langsmith(data: Dict):
+def _log_event_to_langsmith(data: Dict[str, Any]):
+    """
+    Create an Example in LangSmith Dataset.
+    """
     if _ls_client is None:
         return
-
     try:
         inputs = {
             "question": data.get("question", ""),
@@ -134,6 +152,9 @@ def _log_event_to_langsmith(data: Dict):
         print("[langsmith] log failed:", repr(e))
 
 
+# ----------------------------
+# Schemas
+# ----------------------------
 class LoginReq(BaseModel):
     name: str
     user_id: str
@@ -160,17 +181,23 @@ class SummaryReq(BaseModel):
 
 class FeedbackReq(BaseModel):
     user_id: str
-    rating: str
+    rating: str  # "helpful" | "not_helpful"
     assistant_message_id: Optional[str] = None
+
     assistant_text: str
     user_text: Optional[str] = ""
+
     comment: Optional[str] = ""
+
     refs: Optional[List[str]] = []
     learning_mode: Optional[str] = None
     doc_type: Optional[str] = None
     timestamp_ms: Optional[int] = None
 
 
+# ----------------------------
+# API Routes
+# ----------------------------
 @app.post("/api/login")
 def login(req: LoginReq):
     user_id = (req.user_id or "").strip()
@@ -185,9 +212,6 @@ def login(req: LoginReq):
 
 @app.post("/api/chat")
 def chat(req: ChatReq):
-    t0 = time.time()
-    marks = {"start": 0.0}
-
     user_id = (req.user_id or "").strip()
     msg = (req.message or "").strip()
     if not user_id:
@@ -203,25 +227,31 @@ def chat(req: ChatReq):
             ),
             "refs": [],
             "latency_ms": 0.0,
-            "latency_breakdown": {"total_ms": 0.0, "marks_ms": {}, "segments_ms": {}},
         }
 
+    # ----------------------------
+    # 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["language_detect_done"] = (time.time() - t0) * 1000.0
+    marks_ms["language_detect_done"] = (time.time() - t0) * 1000.0
 
+    # weakness update
     sess["weaknesses"] = update_weaknesses_from_message(msg, sess["weaknesses"])
-    marks["weakness_update_done"] = (time.time() - t0) * 1000.0
+    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["cognitive_update_done"] = (time.time() - t0) * 1000.0
+    marks_ms["cognitive_update_done"] = (time.time() - t0) * 1000.0
 
+    # rag retrieve
     rag_context_text, rag_used_chunks = retrieve_relevant_chunks(msg, sess["rag_chunks"])
-    marks["rag_retrieve_done"] = (time.time() - t0) * 1000.0
-
-    rag_context_chars = len(rag_context_text or "")
-    used_chunks_count = len(rag_used_chunks or [])
-    history_len = len(sess.get("history") or [])
+    marks_ms["rag_retrieve_done"] = (time.time() - t0) * 1000.0
 
+    # llm
     try:
         answer, new_history = chat_with_clare(
             message=msg,
@@ -239,9 +269,19 @@ def chat(req: ChatReq):
         print(f"[chat] error: {repr(e)}")
         return JSONResponse({"error": f"chat failed: {repr(e)}"}, status_code=500)
 
-    marks["llm_done"] = (time.time() - t0) * 1000.0
+    marks_ms["llm_done"] = (time.time() - t0) * 1000.0
+    total_ms = marks_ms["llm_done"]
+
+    # segments (delta)
+    ordered = ["start", "language_detect_done", "weakness_update_done", "cognitive_update_done", "rag_retrieve_done", "llm_done"]
+    segments_ms: Dict[str, float] = {}
+    for i in range(1, len(ordered)):
+        a = ordered[i - 1]
+        b = ordered[i]
+        segments_ms[b] = max(0.0, marks_ms.get(b, 0.0) - marks_ms.get(a, 0.0))
+
+    latency_breakdown = {"marks_ms": marks_ms, "segments_ms": segments_ms, "total_ms": total_ms}
 
-    total_ms = (time.time() - t0) * 1000.0
     sess["history"] = new_history
 
     refs = [
@@ -249,31 +289,24 @@ def chat(req: ChatReq):
         for c in (rag_used_chunks or [])
     ]
 
-    # segments_ms：给你一眼看“每段耗时”
-    marks_ms = dict(marks)
-    segments_ms = {}
-    order = ["start", "language_detect_done", "weakness_update_done", "cognitive_update_done", "rag_retrieve_done", "llm_done"]
-    prev = 0.0
-    for k in order[1:]:
-        cur = marks_ms.get(k, prev)
-        segments_ms[k] = max(0.0, cur - prev)
-        prev = cur
-
-    latency_breakdown = {
-        "marks_ms": marks_ms,
-        "segments_ms": segments_ms,
-        "total_ms": total_ms,
-    }
+    # 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
     _log_event_to_langsmith(
         {
-            "experiment_id": "RESP_AI_W10",
+            "experiment_id": EXPERIMENT_ID,
             "student_id": user_id,
             "student_name": sess.get("name", ""),
             "event_type": "chat_turn",
             "timestamp": time.time(),
             "latency_ms": total_ms,
-            "latency_breakdown": json.dumps(latency_breakdown, ensure_ascii=False),
+            "latency_breakdown": latency_breakdown,
+            "rag_context_chars": rag_context_chars,
+            "rag_used_chunks_count": rag_used_chunks_count,
+            "history_len": history_len,
             "question": msg,
             "answer": answer,
             "model_name": sess["model_name"],
@@ -281,9 +314,6 @@ def chat(req: ChatReq):
             "learning_mode": req.learning_mode,
             "doc_type": req.doc_type,
             "refs": refs,
-            "history_len": history_len,
-            "rag_context_chars": rag_context_chars,
-            "rag_used_chunks_count": used_chunks_count,
         }
     )
 
@@ -294,10 +324,6 @@ def chat(req: ChatReq):
         ),
         "refs": refs,
         "latency_ms": total_ms,
-        "latency_breakdown": latency_breakdown,
-        "rag_context_chars": rag_context_chars,
-        "rag_used_chunks_count": used_chunks_count,
-        "history_len": history_len,
     }
 
 
@@ -345,7 +371,7 @@ async def upload(
 
     _log_event_to_langsmith(
         {
-            "experiment_id": "RESP_AI_W10",
+            "experiment_id": EXPERIMENT_ID,
             "student_id": user_id,
             "student_name": sess.get("name", ""),
             "event_type": "upload",
@@ -376,7 +402,7 @@ def api_feedback(req: FeedbackReq):
 
     _log_event_to_langsmith(
         {
-            "experiment_id": "RESP_AI_W10",
+            "experiment_id": EXPERIMENT_ID,
             "student_id": user_id,
             "student_name": student_name,
             "event_type": "feedback",
@@ -437,6 +463,9 @@ def memoryline(user_id: str):
     return {"next_review_label": "T+7", "progress_pct": 0.4}
 
 
+# ----------------------------
+# SPA Fallback
+# ----------------------------
 @app.get("/{full_path:path}")
 def spa_fallback(full_path: str, request: Request):
     if (