gbanyan
43785db595
- Add expert_cache.py: TTL-based in-memory cache (1 hour default) - Add expert_source_service.py: WikidataProvider and ConceptNetProvider - Wikidata SPARQL queries for occupations with Chinese labels - ConceptNet API queries for occupation-related concepts - Random selection from cached pool - Update schemas.py: Add ExpertSource enum (llm/wikidata/conceptnet) - Update ExpertTransformationRequest with expert_source and expert_language - Update router: Conditionally use external sources with LLM fallback - New SSE events: expert_source, expert_fallback - Update frontend types with ExpertSource 🤖 Generated with [Claude Code](https://claude.com/claude-code) Co-Authored-By: Claude <noreply@anthropic.com>
264 lines
11 KiB
Python
264 lines
11 KiB
Python
"""Expert Transformation Agent 路由模組"""
|
||
|
||
import json
|
||
import logging
|
||
from typing import AsyncGenerator, List
|
||
|
||
from fastapi import APIRouter
|
||
from fastapi.responses import StreamingResponse
|
||
|
||
from ..models.schemas import (
|
||
ExpertTransformationRequest,
|
||
ExpertProfile,
|
||
ExpertKeyword,
|
||
ExpertTransformationCategoryResult,
|
||
ExpertTransformationDescription,
|
||
ExpertSource,
|
||
)
|
||
from ..prompts.expert_transformation_prompt import (
|
||
get_expert_generation_prompt,
|
||
get_expert_keyword_generation_prompt,
|
||
get_single_description_prompt,
|
||
)
|
||
from ..services.llm_service import ollama_provider, extract_json_from_response
|
||
from ..services.expert_source_service import expert_source_service
|
||
|
||
logger = logging.getLogger(__name__)
|
||
router = APIRouter(prefix="/api/expert-transformation", tags=["expert-transformation"])
|
||
|
||
|
||
async def generate_expert_transformation_events(
|
||
request: ExpertTransformationRequest,
|
||
all_categories: List[str] # For expert generation context
|
||
) -> AsyncGenerator[str, None]:
|
||
"""Generate SSE events for expert transformation process"""
|
||
try:
|
||
temperature = request.temperature if request.temperature is not None else 0.7
|
||
model = request.model
|
||
|
||
# ========== Step 0: Generate expert team ==========
|
||
yield f"event: expert_start\ndata: {json.dumps({'message': '正在組建專家團隊...', 'source': request.expert_source.value}, ensure_ascii=False)}\n\n"
|
||
|
||
experts: List[ExpertProfile] = []
|
||
actual_source = request.expert_source.value
|
||
|
||
# 決定使用哪種來源生成專家
|
||
use_llm = (
|
||
request.expert_source == ExpertSource.LLM or
|
||
request.custom_experts # 有自訂專家時，使用 LLM 補充
|
||
)
|
||
|
||
if use_llm:
|
||
# LLM 生成專家
|
||
try:
|
||
expert_prompt = get_expert_generation_prompt(
|
||
query=request.query,
|
||
categories=all_categories,
|
||
expert_count=request.expert_count,
|
||
custom_experts=request.custom_experts
|
||
)
|
||
logger.info(f"Expert prompt: {expert_prompt[:200]}")
|
||
|
||
expert_response = await ollama_provider.generate(
|
||
expert_prompt, model=model, temperature=temperature
|
||
)
|
||
logger.info(f"Expert response: {expert_response[:500]}")
|
||
|
||
expert_data = extract_json_from_response(expert_response)
|
||
experts_raw = expert_data.get("experts", [])
|
||
|
||
for exp in experts_raw:
|
||
if isinstance(exp, dict) and all(k in exp for k in ["id", "name", "domain"]):
|
||
experts.append(ExpertProfile(**exp))
|
||
|
||
actual_source = "llm"
|
||
|
||
except Exception as e:
|
||
logger.error(f"Failed to generate experts via LLM: {e}")
|
||
yield f"event: error\ndata: {json.dumps({'error': f'專家團隊生成失敗: {str(e)}'}, ensure_ascii=False)}\n\n"
|
||
return
|
||
else:
|
||
# 外部來源生成專家
|
||
try:
|
||
experts_data, actual_source = await expert_source_service.get_experts(
|
||
source=request.expert_source.value,
|
||
count=request.expert_count,
|
||
language=request.expert_language
|
||
)
|
||
|
||
for i, exp_data in enumerate(experts_data):
|
||
experts.append(ExpertProfile(
|
||
id=f"expert-{i}",
|
||
name=exp_data["name"],
|
||
domain=exp_data["domain"],
|
||
perspective=f"從{exp_data['domain']}角度思考"
|
||
))
|
||
|
||
logger.info(f"Generated {len(experts)} experts from {actual_source}")
|
||
|
||
except Exception as e:
|
||
# 外部來源失敗，fallback 到 LLM
|
||
logger.warning(f"External source failed: {e}, falling back to LLM")
|
||
yield f"event: expert_fallback\ndata: {json.dumps({'original': request.expert_source.value, 'fallback': 'llm', 'reason': str(e)}, ensure_ascii=False)}\n\n"
|
||
|
||
try:
|
||
expert_prompt = get_expert_generation_prompt(
|
||
query=request.query,
|
||
categories=all_categories,
|
||
expert_count=request.expert_count,
|
||
custom_experts=request.custom_experts
|
||
)
|
||
|
||
expert_response = await ollama_provider.generate(
|
||
expert_prompt, model=model, temperature=temperature
|
||
)
|
||
|
||
expert_data = extract_json_from_response(expert_response)
|
||
experts_raw = expert_data.get("experts", [])
|
||
|
||
for exp in experts_raw:
|
||
if isinstance(exp, dict) and all(k in exp for k in ["id", "name", "domain"]):
|
||
experts.append(ExpertProfile(**exp))
|
||
|
||
actual_source = "llm"
|
||
|
||
except Exception as llm_error:
|
||
logger.error(f"LLM fallback also failed: {llm_error}")
|
||
yield f"event: error\ndata: {json.dumps({'error': f'專家團隊生成失敗: {str(llm_error)}'}, ensure_ascii=False)}\n\n"
|
||
return
|
||
|
||
# 回報來源資訊
|
||
yield f"event: expert_source\ndata: {json.dumps({'source': actual_source}, ensure_ascii=False)}\n\n"
|
||
yield f"event: expert_complete\ndata: {json.dumps({'experts': [e.model_dump() for e in experts]}, ensure_ascii=False)}\n\n"
|
||
|
||
if not experts:
|
||
yield f"event: error\ndata: {json.dumps({'error': '無法生成專家團隊'}, ensure_ascii=False)}\n\n"
|
||
return
|
||
|
||
# ========== Step 1: Generate keywords from expert perspectives ==========
|
||
yield f"event: keyword_start\ndata: {json.dumps({'message': f'專家團隊為「{request.category}」的屬性生成關鍵字...'}, ensure_ascii=False)}\n\n"
|
||
|
||
all_expert_keywords: List[ExpertKeyword] = []
|
||
|
||
# For each attribute, ask all experts to generate keywords
|
||
for attr_index, attribute in enumerate(request.attributes):
|
||
try:
|
||
kw_prompt = get_expert_keyword_generation_prompt(
|
||
category=request.category,
|
||
attribute=attribute,
|
||
experts=[e.model_dump() for e in experts],
|
||
keywords_per_expert=request.keywords_per_expert
|
||
)
|
||
logger.info(f"Keyword prompt for '{attribute}': {kw_prompt[:300]}")
|
||
|
||
kw_response = await ollama_provider.generate(
|
||
kw_prompt, model=model, temperature=temperature
|
||
)
|
||
logger.info(f"Keyword response for '{attribute}': {kw_response[:500]}")
|
||
|
||
kw_data = extract_json_from_response(kw_response)
|
||
keywords_raw = kw_data.get("keywords", [])
|
||
|
||
# Add source_attribute to each keyword
|
||
for kw in keywords_raw:
|
||
if isinstance(kw, dict) and all(k in kw for k in ["keyword", "expert_id", "expert_name"]):
|
||
all_expert_keywords.append(ExpertKeyword(
|
||
keyword=kw["keyword"],
|
||
expert_id=kw["expert_id"],
|
||
expert_name=kw["expert_name"],
|
||
source_attribute=attribute
|
||
))
|
||
|
||
# Emit progress
|
||
yield f"event: keyword_progress\ndata: {json.dumps({'attribute': attribute, 'count': len(keywords_raw)}, ensure_ascii=False)}\n\n"
|
||
|
||
except Exception as e:
|
||
logger.warning(f"Failed to generate keywords for '{attribute}': {e}")
|
||
yield f"event: keyword_progress\ndata: {json.dumps({'attribute': attribute, 'count': 0, 'error': str(e)}, ensure_ascii=False)}\n\n"
|
||
# Continue with next attribute instead of stopping
|
||
|
||
yield f"event: keyword_complete\ndata: {json.dumps({'total_keywords': len(all_expert_keywords)}, ensure_ascii=False)}\n\n"
|
||
|
||
if not all_expert_keywords:
|
||
yield f"event: error\ndata: {json.dumps({'error': '無法生成關鍵字'}, ensure_ascii=False)}\n\n"
|
||
return
|
||
|
||
# ========== Step 2: Generate descriptions one by one ==========
|
||
yield f"event: description_start\ndata: {json.dumps({'message': '為專家關鍵字生成創新應用描述...', 'total': len(all_expert_keywords)}, ensure_ascii=False)}\n\n"
|
||
|
||
descriptions: List[ExpertTransformationDescription] = []
|
||
|
||
# Build expert lookup for domain info
|
||
expert_lookup = {exp.id: exp for exp in experts}
|
||
|
||
for idx, kw in enumerate(all_expert_keywords):
|
||
try:
|
||
expert = expert_lookup.get(kw.expert_id)
|
||
expert_domain = expert.domain if expert else ""
|
||
|
||
desc_prompt = get_single_description_prompt(
|
||
query=request.query,
|
||
keyword=kw.keyword,
|
||
expert_id=kw.expert_id,
|
||
expert_name=kw.expert_name,
|
||
expert_domain=expert_domain
|
||
)
|
||
|
||
desc_response = await ollama_provider.generate(
|
||
desc_prompt, model=model, temperature=temperature
|
||
)
|
||
|
||
desc_data = extract_json_from_response(desc_response)
|
||
desc_text = desc_data.get("description", "")
|
||
|
||
if desc_text:
|
||
descriptions.append(ExpertTransformationDescription(
|
||
keyword=kw.keyword,
|
||
expert_id=kw.expert_id,
|
||
expert_name=kw.expert_name,
|
||
description=desc_text
|
||
))
|
||
|
||
# Send progress update
|
||
yield f"event: description_progress\ndata: {json.dumps({'current': idx + 1, 'total': len(all_expert_keywords), 'keyword': kw.keyword}, ensure_ascii=False)}\n\n"
|
||
|
||
except Exception as e:
|
||
logger.warning(f"Failed to generate description for '{kw.keyword}': {e}")
|
||
# Continue with next keyword
|
||
|
||
yield f"event: description_complete\ndata: {json.dumps({'count': len(descriptions)}, ensure_ascii=False)}\n\n"
|
||
|
||
# ========== Build final result ==========
|
||
result = ExpertTransformationCategoryResult(
|
||
category=request.category,
|
||
original_attributes=request.attributes,
|
||
expert_keywords=all_expert_keywords,
|
||
descriptions=descriptions
|
||
)
|
||
|
||
final_data = {
|
||
"result": result.model_dump(),
|
||
"experts": [e.model_dump() for e in experts]
|
||
}
|
||
yield f"event: done\ndata: {json.dumps(final_data, ensure_ascii=False)}\n\n"
|
||
|
||
except Exception as e:
|
||
logger.error(f"Expert transformation error: {e}", exc_info=True)
|
||
yield f"event: error\ndata: {json.dumps({'error': str(e)}, ensure_ascii=False)}\n\n"
|
||
|
||
|
||
@router.post("/category")
|
||
async def expert_transform_category(request: ExpertTransformationRequest):
|
||
"""處理單一類別的專家視角轉換"""
|
||
# Extract all categories from request (should be passed separately in production)
|
||
# For now, use just the single category
|
||
return StreamingResponse(
|
||
generate_expert_transformation_events(request, [request.category]),
|
||
media_type="text/event-stream",
|
||
headers={
|
||
"Cache-Control": "no-cache",
|
||
"Connection": "keep-alive",
|
||
"X-Accel-Buffering": "no",
|
||
},
|
||
)
|