cfc/cfc-langgraph/app/api/indicator_classify.py

from fastapi import APIRouter
from pydantic import BaseModel
from typing import Optional, List
from app.llm.client import get_llm
from langchain_core.messages import SystemMessage, HumanMessage
import logging
import json

logger = logging.getLogger(__name__)
router = APIRouter(prefix="/api/v1", tags=["indicator_classify"])
_llm = None


def get_llm_instance():
    global _llm
    if _llm is None:
        _llm = get_llm()
    return _llm


class ClassifyRequest(BaseModel):
    source_type: str
    raw_name: str
    candidates: Optional[List[str]] = None


class ClassifyResponse(BaseModel):
    code: int = 200
    message: str = "ok"
    data: dict = {}


SYSTEM_PROMPT = (
    "你是健康报告指标归一化归类助手。给定一个报告中的原始指标/菌属/食材名称，"
    "把它归类为统一指标字典中的一个标准定义。只输出 JSON："
    '{"code": "indicator.<类别>.<标准名>" | "bacteria.<标准名>" | "food.<标准名>", '
    '"name": "标准显示名", "category": "分类"}。'
    "若候选列表非空，优先从候选中选择语义最接近的标准 code。"
)


@router.post("/indicator/classify", response_model=ClassifyResponse)
async def classify_indicator(req: ClassifyRequest):
    logger.info("indicator_classify: source=%s raw=%s", req.source_type, req.raw_name)
    llm = get_llm_instance()
    prompt = f"来源类型: {req.source_type}\n原始名称: {req.raw_name}\n"
    if req.candidates:
        prompt += "候选标准定义: " + ", ".join(req.candidates) + "\n"
    try:
        messages = [
            SystemMessage(content=SYSTEM_PROMPT),
            HumanMessage(content=prompt),
        ]
        response = await llm.ainvoke(messages)
        text = response.content.strip()
        if "```json" in text:
            text = text.split("```json")[1].split("```")[0].strip()
        elif "```" in text:
            text = text.split("```")[1].strip()
        data = json.loads(text)
        return ClassifyResponse(data=data)
    except Exception as e:
        logger.error("指标归类失败: %s", e, exc_info=True)
        return ClassifyResponse(code=500, message=f"归类失败: {str(e)}", data={})