add bloom

submission/api.py

@@ -1,3 +1,4 @@
+import threading
 from typing import List, Optional
 from uuid import UUID
 from ninja import Router, Query
@@ -44,7 +45,7 @@ def create_submission(request, payload: SubmissionIn):
         conversation.is_active = False
         conversation.save(update_fields=["is_active"])
 
-    Submission.objects.create(
+    sub = Submission.objects.create(
         user=request.user,
         task=task,
         html=payload.html,
@@ -53,6 +54,10 @@ def create_submission(request, payload: SubmissionIn):
         conversation=conversation,
     )
 
+    if conversation:
+        from .classifier import classify_conversation_messages
+        threading.Thread(target=classify_conversation_messages, args=(conversation.id,), daemon=True).start()
+
 
 @router.get("/", response=List[SubmissionOut])
 @paginate
@@ -360,3 +365,5 @@ def update_flag(request, submission_id: UUID, payload: FlagIn):
     return {"flag": submission.flag}
 
 
+
+

submission/classifier.py

@@ -0,0 +1,86 @@
+import re
+import time
+import logging
+from uuid import UUID
+
+from django.conf import settings
+from openai import OpenAI
+
+logger = logging.getLogger(__name__)
+
+CLASSIFY_SYSTEM_PROMPT = """你是一个教育评估专家。根据布鲁姆认知分类学，分析以下学生在前端学习中发送给AI助手的一条提示词，判断该提示词所体现的认知层级。
+
+层级定义：
+- L1 记忆：能背诵HTML标签语法（例："帮我写一个按钮"）
+- L2 理解：能解释flex布局原理（例："为什么这里不居中？"）
+- L3 应用：能独立搭建页面结构（例："用flex做导航栏，间距16px"）
+- L4 分析：能定位跨浏览器兼容性bug（例："Safari中margin失效，原因？"）
+- L5 评价：能对比并选择方案（例："对比Grid与Flex方案优劣"）
+- L6 创造：能设计并实现原创交互作品（例："设计夜间/日间切换效果"）
+
+只返回一个数字（1-6），不要解释。"""
+
+
+def _call_llm(content: str) -> int | None:
+    """Call LLM to classify a single message content. Returns level 1-6 or None."""
+    try:
+        client = OpenAI(
+            api_key=settings.LLM_API_KEY,
+            base_url=settings.LLM_BASE_URL,
+            timeout=30.0,
+        )
+        response = client.chat.completions.create(
+            model=settings.LLM_MODEL,
+            messages=[
+                {"role": "system", "content": CLASSIFY_SYSTEM_PROMPT},
+                {"role": "user", "content": content},
+            ],
+            max_tokens=10,
+            stream=False,
+        )
+        text = response.choices[0].message.content or ""
+        match = re.search(r"[1-6]", text)
+        if not match:
+            logger.warning("classify: unexpected LLM response '%s'", text)
+            return None
+        return int(match.group())
+    except Exception as e:
+        logger.error("classify LLM call failed: %s", e)
+        return None
+
+
+def classify_message(message_id: int) -> int | None:
+    """Classify a single user Message by ID. Returns level or None."""
+    from prompt.models import Message
+
+    try:
+        msg = Message.objects.get(id=message_id, role="user")
+    except Message.DoesNotExist:
+        return None
+
+    level = _call_llm(msg.content)
+    if level is not None:
+        Message.objects.filter(id=message_id).update(prompt_level=level)
+    return level
+
+
+def classify_conversation_messages(conversation_id: UUID, force: bool = False) -> None:
+    """Classify all user messages in a conversation."""
+    from prompt.models import Message
+
+    qs = Message.objects.filter(conversation_id=conversation_id, role="user")
+    if not force:
+        qs = qs.filter(prompt_level__isnull=True)
+
+    for msg in qs.order_by("created"):
+        level = _call_llm(msg.content)
+        if level is not None:
+            Message.objects.filter(id=msg.id).update(prompt_level=level)
+        time.sleep(0.3)
+
+
+def classify_messages_batch(message_ids: list) -> None:
+    """Classify a list of messages by ID."""
+    for mid in message_ids:
+        classify_message(mid)
+        time.sleep(0.5)

submission/management/commands/classify_prompts.py

@@ -0,0 +1,35 @@
+from django.core.management.base import BaseCommand
+
+from prompt.models import Message
+from submission.classifier import classify_message
+
+
+class Command(BaseCommand):
+    help = "Classify prompt levels (L1-L6) for user messages using LLM"
+
+    def add_arguments(self, parser):
+        parser.add_argument("--task-id", type=int, help="Only classify messages for this task ID")
+        parser.add_argument("--force", action="store_true", help="Re-classify already classified messages")
+        parser.add_argument("--dry-run", action="store_true", help="Show count without classifying")
+
+    def handle(self, *args, **options):
+        qs = Message.objects.filter(role="user")
+        if options["task_id"]:
+            qs = qs.filter(conversation__task_id=options["task_id"])
+        if not options["force"]:
+            qs = qs.filter(prompt_level__isnull=True)
+
+        ids = list(qs.values_list("id", flat=True))
+        self.stdout.write(f"Found {len(ids)} message(s) to classify.")
+
+        if options["dry_run"]:
+            self.stdout.write("Dry run — no changes made.")
+            return
+
+        for i, mid in enumerate(ids, 1):
+            level = classify_message(mid)
+            self.stdout.write(
+                f"[{i}/{len(ids)}] msg#{mid} → L{level}" if level else f"[{i}/{len(ids)}] msg#{mid} → (skipped)"
+            )
+
+        self.stdout.write(self.style.SUCCESS("Done."))