0063df0a87
- Бэкенд: FastAPI + SQLAlchemy async + PostgreSQL, JWT-аутентификация, Qdrant multi-tenant векторное хранилище, Celery воркер - Фронтенд: React + Vite + TypeScript SPA (логин, регистрация, дашборд, поиск по коллекциям, настройки) - Инфраструктура: docker-compose, Helm-чарт, GitHub Actions CI, Prometheus/Grafana мониторинг - Embedding через OpenRouter (nvidia/llama-nemotron-embed-vl-1b-v2:free) - LLM: google/gemma-4-31b-it:free через OpenRouter
69 lines
2.1 KiB
Python
69 lines
2.1 KiB
Python
import uuid
|
|
|
|
from fastapi import APIRouter, Depends, HTTPException, status
|
|
from sqlalchemy import select
|
|
from sqlalchemy.ext.asyncio import AsyncSession
|
|
|
|
from app.auth.deps import get_current_user
|
|
from app.database import get_db
|
|
from app.models.collection import Collection
|
|
from app.models.search_schemas import SearchRequest, SearchResponse, SourceItem
|
|
from app.models.user import User
|
|
from app.services.embedding import embed_query
|
|
from app.services.vector_store import search as qdrant_search
|
|
from app.services.reranking import rerank
|
|
from app.services.llm import generate_answer
|
|
|
|
router = APIRouter(prefix="/api", tags=["search"])
|
|
|
|
FETCH_K = 20
|
|
FINAL_K = 5
|
|
|
|
|
|
@router.post("/search", response_model=SearchResponse)
|
|
async def search(
|
|
body: SearchRequest,
|
|
user: User = Depends(get_current_user),
|
|
db: AsyncSession = Depends(get_db),
|
|
):
|
|
result = await db.execute(
|
|
select(Collection).where(
|
|
Collection.id == uuid.UUID(body.collection_id),
|
|
Collection.user_id == user.id,
|
|
)
|
|
)
|
|
collection = result.scalar_one_or_none()
|
|
if not collection:
|
|
raise HTTPException(status_code=status.HTTP_404_NOT_FOUND, detail="Collection not found")
|
|
|
|
query_vector = await embed_query(body.query)
|
|
|
|
qdrant_results = await qdrant_search(
|
|
tenant_id=str(user.id),
|
|
query_vector=query_vector,
|
|
collection_id=body.collection_id,
|
|
k=FETCH_K,
|
|
)
|
|
|
|
if not qdrant_results:
|
|
return SearchResponse(
|
|
answer="Недостаточно информации в истории обращений" if body.generate_answer else None,
|
|
sources=[],
|
|
)
|
|
|
|
reranked = rerank(body.query, qdrant_results, k=FINAL_K)
|
|
|
|
sources = [
|
|
SourceItem(ticket_id=r["ticket_id"], score=r["score"], category=r["category"])
|
|
for r in reranked
|
|
]
|
|
|
|
answer = None
|
|
if body.generate_answer:
|
|
context = "\n\n".join(
|
|
f"Тикет №{r['ticket_id']} ({r['category']}):\n{r['full_text']}"
|
|
for r in reranked
|
|
)
|
|
answer = await generate_answer(body.query, context)
|
|
|
|
return SearchResponse(answer=answer, sources=sources)
|