import os from app.agents.base_agent import BaseAgent MOCK_RESULT = { "must_have": [ "Java", "SpringBoot", "MyBatis", "MySQL", "Redis", "分布式系统", "微服务架构", "本科及以上学历", "计算机相关专业", ], "nice_to_have": [ "Kubernetes", "Docker", "消息队列", "Elasticsearch", "大流量高并发经验", "开源项目贡献", ], "buzzwords": [ "具有较强的学习能力和团队协作精神", "工作积极主动,责任心强", "具备良好的沟通表达能力", "有较强的抗压能力", ], "expectation_signals": [ {"tag": "加班文化", "evidence": "能承受较大工作压力", "confidence": "medium"}, {"tag": "技术氛围", "evidence": "钻研前沿技术,追求卓越", "confidence": "high"}, {"tag": "团队协作", "evidence": "具备良好的团队协作精神", "confidence": "medium"}, {"tag": "结果导向", "evidence": "以结果为导向,注重交付质量", "confidence": "high"}, ], "position_summary": "Java后端开发工程师,负责核心业务系统的设计与开发,要求具备扎实的Java基础和微服务实践经验", } class JdParser(BaseAgent): name = "jd_parser" def build_prompt(self, jd_text: str) -> tuple[str, dict]: system = ( "你是一个JD解析专家。请对提供的JD文本执行以下四项任务:\n\n" "1. 去噪(Denoise):识别并过滤掉模板话术/通用套话(buzzwords)," "这些描述通常在任何岗位中都能出现,缺乏具体信息量。\n\n" "2. 分层(Layer):将JD中要求的能力/条件分为两类:\n" " - must_have:明确要求的必备技能或条件\n" " - nice_to_have:加分项、优先考虑项、期望但不强制\n\n" "3. 翻译(Translate):将模糊描述转换为具体的技术栈或技能标签。\n" " 例如:'熟悉主流框架' → SpringBoot, MyBatis\n" " 例如:'熟悉数据库技术' → MySQL, Redis\n" " 例如:'熟练掌握Linux操作' → Linux\n\n" "4. 期望信号提取(Expectation Signals):提取JD中隐含的关于加班文化、" "团队氛围、薪酬待遇、技术风格等信号。\n" " 每条信号必须包含:\n" " - tag:信号类别(如'加班文化''技术氛围''结果导向''团队协作'等)\n" " - evidence:原文引用(精确的JD原文段落)\n" " - confidence:置信度(high/medium/low)\n\n" "请严格按以下JSON格式输出:\n" "{\n" ' "must_have": ["技能1", "技能2", ...],\n' ' "nice_to_have": ["技能1", "技能2", ...],\n' ' "buzzwords": ["套话1", "套话2", ...],\n' ' "expectation_signals": [\n' ' {"tag": "信号类别", "evidence": "原文引用", "confidence": "high/medium/low"}\n' " ],\n" ' "position_summary": "一句话总结该岗位"\n' "}" ) user = {"jd_text": jd_text} return system, user async def parse_jd(self, jd_text: str) -> dict: if not os.getenv("DEEPSEEK_API_KEY"): return dict(MOCK_RESULT) system, user_input = self.build_prompt(jd_text) result = await self.call_llm(system, user_input) return result