Update api/server.py

api/server.py

@@ -1,7 +1,7 @@
 # api/server.py
 import os
 import time
-from typing import Dict
+from typing import Dict, Any, Optional, List
 
 from fastapi import FastAPI, UploadFile, File, Form, Request
 from fastapi.responses import FileResponse, JSONResponse
@@ -22,6 +22,58 @@ from api.clare_core import (
     summarize_conversation,
 )
 
+# ----------------------------
+# LangSmith (Dataset logging)
+# ----------------------------
+# 你在 HF Space 里需要配置：
+#   LANGSMITH_API_KEY=...
+# 可选：
+#   LANGSMITH_DATASET_NAME=clare_user_events
+#   LANGSMITH_PROJECT=...
+try:
+    from langsmith import Client as LangSmithClient  # type: ignore
+except Exception:
+    LangSmithClient = None  # type: ignore
+
+LS_DATASET_NAME = os.getenv("LANGSMITH_DATASET_NAME", "clare_user_events").strip()
+LS_PROJECT = os.getenv("LANGSMITH_PROJECT", "").strip()
+
+_ls_client = None
+if LangSmithClient is not None and os.getenv("LANGSMITH_API_KEY"):
+    try:
+        _ls_client = LangSmithClient()
+    except Exception as e:
+        print(f"[langsmith] init failed: {repr(e)}")
+        _ls_client = None
+
+
+def log_event_to_langsmith(
+    *,
+    inputs: Dict[str, Any],
+    outputs: Dict[str, Any],
+    metadata: Dict[str, Any],
+) -> None:
+    """
+    Write a single event as an Example row into LangSmith Dataset.
+    This mirrors your old Gradio pattern (dataset作为事件日志).
+    """
+    if _ls_client is None:
+        return
+    try:
+        # project 不是必须；dataset 足够你做过滤与分析
+        if LS_PROJECT:
+            metadata = {**metadata, "langsmith_project": LS_PROJECT}
+
+        _ls_client.create_example(
+            inputs=inputs,
+            outputs=outputs,
+            metadata=metadata,
+            dataset_name=LS_DATASET_NAME,
+        )
+    except Exception as e:
+        print(f"[langsmith] create_example failed: {repr(e)}")
+
+
 # ----------------------------
 # Paths / Constants
 # ----------------------------
@@ -40,7 +92,6 @@ WEB_ASSETS = os.path.join(WEB_DIST, "assets")
 # ----------------------------
 app = FastAPI(title="Clare API")
 
-# Same-origin for Docker Space doesn't need CORS, but leaving it open helps if you later split FE/BE.
 app.add_middleware(
     CORSMiddleware,
     allow_origins=["*"],
@@ -52,11 +103,9 @@ app.add_middleware(
 # ----------------------------
 # Static hosting (Vite build)
 # ----------------------------
-# Mount /assets so <script src="/assets/..."> works.
 if os.path.isdir(WEB_ASSETS):
     app.mount("/assets", StaticFiles(directory=WEB_ASSETS), name="assets")
 
-# Optional: serve other static files in build root (e.g., favicon) under /static
 if os.path.isdir(WEB_DIST):
     app.mount("/static", StaticFiles(directory=WEB_DIST), name="static")
 
@@ -99,11 +148,8 @@ def _get_session(user_id: str) -> Dict:
             "weaknesses": [],
             "cognitive_state": {"confusion": 0, "mastery": 0},
             "course_outline": DEFAULT_COURSE_TOPICS,
-            # preload base reading
             "rag_chunks": list(MODULE10_CHUNKS_CACHE),
             "model_name": DEFAULT_MODEL,
-            # ✅ NEW: track last syllabus filename for refs fallback
-            "last_syllabus_file": None,
         }
     return SESSIONS[user_id]
 
@@ -135,6 +181,22 @@ class SummaryReq(BaseModel):
     language_preference: str = "Auto"
 
 
+class FeedbackReq(BaseModel):
+    # FE 会发的最小字段
+    user_id: str
+    rating: str  # "helpful" | "not_helpful"
+    assistant_message_id: str
+    assistant_text: str
+
+    # 可选：用于更好的分析
+    user_text: Optional[str] = None
+    comment: Optional[str] = None
+    refs: Optional[List[str]] = None
+    learning_mode: Optional[str] = None
+    doc_type: Optional[str] = None
+    timestamp_ms: Optional[float] = None
+
+
 # ----------------------------
 # API Routes
 # ----------------------------
@@ -175,7 +237,6 @@ def chat(req: ChatReq):
     sess["weaknesses"] = update_weaknesses_from_message(msg, sess["weaknesses"])
     sess["cognitive_state"] = update_cognitive_state_from_message(msg, sess["cognitive_state"])
 
-    # RAG
     rag_context_text, rag_used_chunks = retrieve_relevant_chunks(msg, sess["rag_chunks"])
 
     start_ts = time.time()
@@ -204,24 +265,27 @@ def chat(req: ChatReq):
         for c in (rag_used_chunks or [])
     ]
 
-    # ✅ NEW: Syllabus chat should show syllabus in refs (fallback if retrieval didn't hit it)
-    doc_type_norm = (req.doc_type or "").strip().lower()
-    if doc_type_norm == "syllabus":
-        syllabus_name = sess.get("last_syllabus_file") or "uploaded_syllabus"
-        # We consider it a hit if any ref source_file contains the syllabus filename
-        def _has_syllabus_ref() -> bool:
-            for r in refs:
-                sf = (r.get("source_file") or "").strip()
-                if not sf:
-                    continue
-                # robust matching: exact basename containment OR endswith
-                if syllabus_name in sf or sf.endswith(syllabus_name):
-                    return True
-            return False
-
-        if not _has_syllabus_ref():
-            # put syllabus at the front, preserve existing refs after
-            refs = [{"source_file": syllabus_name, "section": "syllabus_outline"}] + refs
+    # 可选：把 chat_turn 也写入 dataset（你以前 Gradio 有）
+    try:
+        log_event_to_langsmith(
+            inputs={
+                "question": msg,
+                "student_id": user_id,
+            },
+            outputs={
+                "answer": answer,
+            },
+            metadata={
+                "event_type": "chat_turn",
+                "timestamp": time.time(),
+                "latency_ms": latency_ms,
+                "learning_mode": req.learning_mode,
+                "language": resolved_lang,
+                "doc_type": req.doc_type,
+            },
+        )
+    except Exception:
+        pass
 
     return {
         "reply": answer,
@@ -249,7 +313,6 @@ async def upload(
 
     sess = _get_session(user_id)
 
-    # Save to /tmp (sanitize filename)
     safe_name = os.path.basename(file.filename).replace("..", "_")
     tmp_path = os.path.join("/tmp", safe_name)
 
@@ -257,11 +320,6 @@ async def upload(
     with open(tmp_path, "wb") as f:
         f.write(content)
 
-    # ✅ NEW: remember the latest syllabus filename for refs fallback
-    if doc_type == "Syllabus":
-        sess["last_syllabus_file"] = os.path.basename(file.filename) or safe_name
-
-    # Update topics only for syllabus
     if doc_type == "Syllabus":
         class _F:
             pass
@@ -273,7 +331,6 @@ async def upload(
         except Exception as e:
             print(f"[upload] syllabus parse error: {repr(e)}")
 
-    # Update rag chunks for any doc
     try:
         new_chunks = build_rag_chunks_from_file(tmp_path, doc_type) or []
         sess["rag_chunks"] = (sess["rag_chunks"] or []) + new_chunks
@@ -285,6 +342,43 @@ async def upload(
     return {"ok": True, "added_chunks": len(new_chunks), "status_md": status_md}
 
 
+@app.post("/api/feedback")
+def feedback(req: FeedbackReq):
+    user_id = (req.user_id or "").strip()
+    if not user_id:
+        return JSONResponse({"ok": False, "error": "Missing user_id"}, status_code=400)
+
+    rating = (req.rating or "").strip().lower()
+    if rating not in ("helpful", "not_helpful"):
+        return JSONResponse({"ok": False, "error": "rating must be helpful|not_helpful"}, status_code=400)
+
+    # 写入 LangSmith dataset（与你 Gradio 时代一致）
+    try:
+        log_event_to_langsmith(
+            inputs={
+                "question": req.user_text or "",   # 允许为空（只对 assistant reply 点赞）
+                "student_id": user_id,
+                "assistant_message_id": req.assistant_message_id,
+            },
+            outputs={
+                "answer": req.assistant_text or "",
+            },
+            metadata={
+                "event_type": "feedback",
+                "rating": rating,
+                "comment": (req.comment or "").strip(),
+                "learning_mode": req.learning_mode or "",
+                "doc_type": req.doc_type or "",
+                "refs": req.refs or [],
+                "timestamp_ms": req.timestamp_ms or (time.time() * 1000.0),
+            },
+        )
+    except Exception as e:
+        print(f"[feedback] log failed: {repr(e)}")
+
+    return {"ok": True}
+
+
 @app.post("/api/export")
 def api_export(req: ExportReq):
     user_id = (req.user_id or "").strip()
@@ -323,17 +417,14 @@ def api_summary(req: SummaryReq):
 @app.get("/api/memoryline")
 def memoryline(user_id: str):
     _ = _get_session((user_id or "").strip())
-    # v1: 写死也没问题；前端只渲染
     return {"next_review_label": "T+7", "progress_pct": 0.4}
 
 
 # ----------------------------
-# SPA Fallback (important!)
+# SPA Fallback
 # ----------------------------
-# If user refreshes /some/route, FE router needs index.html.
 @app.get("/{full_path:path}")
 def spa_fallback(full_path: str, request: Request):
-    # Do not hijack API/static paths
     if (
         full_path.startswith("api/")
         or full_path.startswith("assets/")