feat: RAG 임베딩 자동 동기화 + AI 서비스 개선

- 부적합 라이프사이클 전 과정에서 Qdrant 임베딩 자동 동기화
  - 관리함 5개 저장 함수 + 수신함 상태 변경 시 fire-and-forget sync
  - 30분 주기 전체 재동기화 안전망 (FastAPI lifespan 백그라운드 태스크)
  - build_document_text에 카테고리(final_category/category) 포함
- RAG 질의에 DB 통계 집계 지원 (카테고리별/부서별 건수)
- Qdrant client.search → query_points API 마이그레이션
- AI 어시스턴트 페이지 권한 추가 (tkuser)

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

ai-service/services/db_client.py

@@ -82,6 +82,38 @@ def get_daily_qc_stats(date_str: str) -> dict:
         return dict(row._mapping) if row else {}
 
 
+def get_category_stats() -> list[dict]:
+    """카테고리별 부적합 건수 집계"""
+    with engine.connect() as conn:
+        result = conn.execute(
+            text(
+                "SELECT COALESCE(final_category, category) AS category, "
+                "COUNT(*) AS count "
+                "FROM qc_issues "
+                "GROUP BY COALESCE(final_category, category) "
+                "ORDER BY count DESC"
+            )
+        )
+        return [dict(row._mapping) for row in result]
+
+
+def get_department_stats() -> list[dict]:
+    """부서별 부적합 건수 집계"""
+    with engine.connect() as conn:
+        result = conn.execute(
+            text(
+                "SELECT responsible_department AS department, "
+                "COUNT(*) AS count "
+                "FROM qc_issues "
+                "WHERE responsible_department IS NOT NULL "
+                "AND responsible_department != '' "
+                "GROUP BY responsible_department "
+                "ORDER BY count DESC"
+            )
+        )
+        return [dict(row._mapping) for row in result]
+
+
 def get_issues_for_date(date_str: str) -> list[dict]:
     with engine.connect() as conn:
         result = conn.execute(

ai-service/services/embedding_service.py

@@ -1,11 +1,18 @@
+import logging
+
 from services.ollama_client import ollama_client
 from db.vector_store import vector_store
 from db.metadata_store import metadata_store
 from services.db_client import get_all_issues, get_issue_by_id, get_issues_since
 
+logger = logging.getLogger(__name__)
+
 
 def build_document_text(issue: dict) -> str:
     parts = []
+    cat = issue.get("final_category") or issue.get("category")
+    if cat:
+        parts.append(f"분류: {cat}")
     if issue.get("description"):
         parts.append(issue["description"])
     if issue.get("final_description"):
@@ -84,6 +91,7 @@ async def sync_all_issues() -> dict:
 
 
 async def sync_single_issue(issue_id: int) -> dict:
+    logger.info(f"Sync single issue: {issue_id}")
     issue = get_issue_by_id(issue_id)
     if not issue:
         return {"status": "not_found"}

ai-service/services/rag_service.py

@@ -1,8 +1,54 @@
+import logging
+import time
+
 from services.ollama_client import ollama_client
 from services.embedding_service import search_similar_by_text, build_document_text
-from services.db_client import get_issue_by_id
+from services.db_client import get_issue_by_id, get_category_stats, get_department_stats
 from services.utils import load_prompt
 
+logger = logging.getLogger(__name__)
+
+_stats_cache = {"data": "", "expires": 0}
+STATS_CACHE_TTL = 300  # 5분
+
+STATS_KEYWORDS = {"많이", "빈도", "추이", "비율", "통계", "몇 건", "자주", "빈번", "유형별", "부서별"}
+
+
+def _needs_stats(question: str) -> bool:
+    """키워드 매칭으로 통계성 질문인지 판별"""
+    return any(kw in question for kw in STATS_KEYWORDS)
+
+
+def _build_stats_summary() -> str:
+    """DB 집계 통계 요약 (5분 TTL 캐싱, 실패 시 빈 문자열)"""
+    now = time.time()
+    if _stats_cache["data"] and now < _stats_cache["expires"]:
+        return _stats_cache["data"]
+    try:
+        lines = ["[전체 통계 요약]"]
+        cats = get_category_stats()
+        if cats:
+            total = sum(c["count"] for c in cats)
+            lines.append(f"총 부적합 건수: {total}건")
+            lines.append("카테고리별:")
+            for c in cats[:10]:
+                pct = round(c["count"] / total * 100, 1)
+                lines.append(f"  - {c['category']}: {c['count']}건 ({pct}%)")
+        depts = get_department_stats()
+        if depts:
+            lines.append("부서별:")
+            for d in depts[:10]:
+                lines.append(f"  - {d['department']}: {d['count']}건")
+        if len(lines) <= 1:
+            return ""  # 데이터 없으면 빈 문자열
+        result = "\n".join(lines)
+        _stats_cache["data"] = result
+        _stats_cache["expires"] = now + STATS_CACHE_TTL
+        return result
+    except Exception as e:
+        logger.warning(f"Stats summary failed: {e}")
+        return ""
+
 
 def _format_retrieved_issues(results: list[dict]) -> str:
     if not results:
@@ -81,11 +127,16 @@ async def rag_ask(question: str, project_id: int = None) -> dict:
     results = await search_similar_by_text(
         question, n_results=7, filters=None
     )
+    logger.info(f"RAG ask: question='{question[:50]}', results={len(results)}")
     context = _format_retrieved_issues(results)
 
+    # 통계성 질문일 때만 DB 집계 포함 (토큰 절약)
+    stats = _build_stats_summary() if _needs_stats(question) else ""
+
     template = load_prompt("prompts/rag_qa.txt")
     prompt = template.format(
         question=question,
+        stats_summary=stats,
         retrieved_cases=context,
     )