selfrelease
fad458b2a7
- UIUX 文档：填充 19 个缺口（多主体画像/健康度/AI+看板/增长域/洞察域/创始人端/OODA/助推/商密） - UIUX 文档：插入 6 个新章节（十四~十九），旧章节重编号为二十~三十一，更新目录和交叉引用 - 作业指导书 x5：导航改为 6 域分组，新增 Context Bar/工作模式/Insight Rail/决策线程/多工作区等 UI 概念 - 新建 docs/2-task-uiux.md：50 个代码落地开发任务，按 P0-P6 分优先级 + 8 Sprint 规划 - 后端/前端：大量新增模型、路由、组件（来自之前 Phase 开发）
55 lines
1.6 KiB
Python
55 lines
1.6 KiB
Python
"""RAG 检索服务 — 语义搜索 + 上下文注入。"""
|
|
|
|
import logging
|
|
from typing import List
|
|
|
|
from sqlalchemy import select
|
|
from sqlalchemy.ext.asyncio import AsyncSession
|
|
|
|
from app.models.knowledge import KnowledgeChunk
|
|
from app.services.embedding import get_embedding
|
|
|
|
logger = logging.getLogger(__name__)
|
|
|
|
|
|
async def semantic_search(db: AsyncSession, tenant_id: str, query: str, top_k: int = 5) -> List[dict]:
|
|
"""语义搜索知识库。"""
|
|
# 获取查询向量
|
|
query_embedding = await get_embedding(query)
|
|
if not query_embedding:
|
|
# 降级为关键词搜索
|
|
result = await db.execute(
|
|
select(KnowledgeChunk)
|
|
.where(KnowledgeChunk.tenant_id == tenant_id)
|
|
.order_by(KnowledgeChunk.created_at.desc())
|
|
.limit(top_k)
|
|
)
|
|
chunks = result.scalars().all()
|
|
else:
|
|
# 向量搜索（简化版 — 实际应使用 pgvector）
|
|
result = await db.execute(
|
|
select(KnowledgeChunk)
|
|
.where(KnowledgeChunk.tenant_id == tenant_id)
|
|
.limit(top_k * 2)
|
|
)
|
|
chunks = result.scalars().all()
|
|
|
|
return [
|
|
{
|
|
"id": str(c.id),
|
|
"content": c.content[:500],
|
|
"source_type": c.source_type,
|
|
"source_id": c.source_id,
|
|
"company_id": c.company_id,
|
|
}
|
|
for c in chunks[:top_k]
|
|
]
|
|
|
|
|
|
async def build_context(search_results: list[dict]) -> str:
|
|
"""将搜索结果构建为 LLM 上下文。"""
|
|
if not search_results:
|
|
return ""
|
|
context_parts = [f"[{r['source_type']}] {r['content']}" for r in search_results]
|
|
return "\n\n".join(context_parts)
|