import os from copy import deepcopy from app.agents.base_agent import BaseAgent _MOCK_RESULT = { "dimensions": [ {"name": "技能匹配", "score": 80, "evidence": {"jd_quote": "熟悉Python", "resume_quote": "熟练使用Python进行数据处理和模型开发"}}, {"name": "经验匹配", "score": 40, "evidence": {"jd_quote": "3年以上相关经验", "resume_quote": "6个月实习经历"}}, {"name": "学历匹配", "score": 100, "evidence": {"jd_quote": "计算机相关专业", "resume_quote": "计算机科学与技术专业"}}, {"name": "项目契合", "score": 65, "evidence": {"jd_quote": "有推荐系统经验优先", "resume_quote": "课程项目实现了协同过滤推荐算法"}}, {"name": "软性素质", "score": 75, "evidence": {"jd_quote": "具备团队协作能力", "resume_quote": "项目描述中提及与3名同学协作完成"}}, ], "total_score": 72, "strength": "技能基础扎实,学历完全符合要求", "gap": "缺乏全职工作经验,项目深度有待提升", "hr_view": { "first_impression": "技能列表与 JD 匹配度高,但项目描述缺乏量化成果", "attention_points": [ {"aspect": "技能栈匹配", "finding": "Python/Flask 完全覆盖 JD 要求", "concern": "无"}, {"aspect": "项目深度", "finding": "仅课程项目,无企业级实践", "concern": "项目描述偏功能实现,未见性能优化或架构设计"}, {"aspect": "学历背景", "finding": "985 计算机专业,符合门槛", "concern": "无"}, ], "hr_concern_level": "medium", "screen_summary": "简历整体合格,项目深度是最大短板,建议补充量化成果", }, } class Matcher(BaseAgent): name = "matcher" def build_prompt(self, jd_parsed: dict, resume_data: dict) -> tuple[str, dict]: system = ( "你是一个简历匹配诊断专家。你的任务是将候选人的简历与岗位要求进行多维度匹配评估。\n\n" "匹配维度(固定5个):\n" "1. 技能匹配 — 候选人掌握的技术栈 vs JD要求的技术栈\n" "2. 经验匹配 — 工作/实习年限 vs JD要求\n" "3. 学历匹配 — 学历/专业 vs JD要求\n" "4. 项目契合 — 项目经历与岗位方向的相关度\n" "5. 软性素质 — 沟通/协作/自驱力(从项目描述中推断)\n\n" "评分规则:\n" "- 每个维度 0-100 分\n" "- 每个分数必须附 evidence 字段:\n" " - jd_quote: JD中的原文依据\n" " - resume_quote: 简历中的原文依据\n" "- 证据必须精确到原文引用,不能模糊概括\n\n" "额外要求:\n" "- total_score 为5个维度的加权总分(权重:技能0.25, 经验0.20, 学历0.15, 项目0.25, 软性0.15)\n" "- strength 总结候选人优势\n" "- gap 总结候选人差距\n" "- hr_view 字段提供HR快速筛查模拟视角,包含 first_impression / attention_points(aspect/finding/concern)/ hr_concern_level(low/medium/high)/ screen_summary\n\n" "请严格按以下JSON格式输出,字段名必须完全匹配:\n" '{\n' ' "dimensions": [\n' ' {"name": "技能匹配", "score": 80, "evidence": {"jd_quote": "JD原文", "resume_quote": "简历原文"}},\n' ' {"name": "经验匹配", "score": 40, "evidence": {"jd_quote": "...", "resume_quote": "..."}},\n' ' {"name": "学历匹配", "score": 100, "evidence": {"jd_quote": "...", "resume_quote": "..."}},\n' ' {"name": "项目契合", "score": 65, "evidence": {"jd_quote": "...", "resume_quote": "..."}},\n' ' {"name": "软性素质", "score": 75, "evidence": {"jd_quote": "...", "resume_quote": "..."}}\n' ' ],\n' ' "total_score": 72,\n' ' "strength": "优势总结",\n' ' "gap": "差距总结",\n' ' "hr_view": {\n' ' "first_impression": "...",\n' ' "attention_points": [\n' ' {"aspect": "...", "finding": "...", "concern": "..."}\n' ' ],\n' ' "hr_concern_level": "medium",\n' ' "screen_summary": "..."\n' ' }\n' '}\n\n' "确保 total_score 为 0-100 的整数(加权:技能0.25, 经验0.20, 学历0.15, 项目0.25, 软性0.15)。\n\n" "输出格式: JSON" ) user = {"jd_parsed": jd_parsed, "resume_data": resume_data} return system, user async def match_resume(self, resume_data: dict, jd_parsed: dict) -> dict: api_key = os.environ.get("DEEPSEEK_API_KEY") if not api_key or os.environ.get("MOCK_MODE", "").lower() == "true": return deepcopy(_MOCK_RESULT) system, user = self.build_prompt(jd_parsed=jd_parsed, resume_data=resume_data) result = await self.call_llm(system, user) raw_dims = result.get("dimensions") or result.get("dimension_list") or result.get("matches") or [] normalized = [] for d in raw_dims: if isinstance(d, dict): normalized.append({ "name": d.get("name") or d.get("dimension") or d.get("title") or "", "score": d.get("score") or d.get("value") or 0, "evidence": d.get("evidence") or d.get("detail") or {"jd_quote": "", "resume_quote": ""}, }) result["dimensions"] = normalized if "error" in result: return deepcopy(_MOCK_RESULT) return result