import os from app.agents.base_agent import BaseAgent MOCK_RESPONSE = { "rewrite_suggestions": [ { "dimension": "技能匹配", "issue": "JD要求Python后端开发经验,简历未体现", "original": "熟悉Java、C++,了解数据结构与算法", "suggestion": "补充Python后端项目经历,例如\"使用FastAPI开发RESTful API,实现用户认证与数据持久化\"", "reason": "JD核心技术栈为Python后端,仅列举Java/C++无法满足岗位核心要求,需明确展示Python实战经验" }, { "dimension": "经验匹配", "issue": "缺少企业级后端实习经历", "original": "在校课程项目:图书管理系统", "suggestion": "突出项目难点与高并发场景,例如\"优化数据库查询减少50%响应时间,支撑日均10万请求\"", "reason": "JD要求有后端实习经验,课程项目需用量化成果弥补实习缺失,展示工程化思维" }, { "dimension": "项目契合", "issue": "项目方向与JD业务场景不匹配", "original": "基于Java Swing的桌面应用开发", "suggestion": "将项目描述重构为\"基于Python的Web全栈项目\",体现完整开发流程", "reason": "JD业务涉及Web后端服务,桌面应用项目与岗位技术栈差异大,需重新包装体现相关能力" }, ], "interview_tips": [ { "topic": "Python异步编程与FastAPI", "reason": "JD明确要求asyncio和FastAPI,简历中无相关经验", "tip": "准备asyncio核心概念(事件循环、协程、Task),结合FastAPI源码理解异步请求处理流程,准备一个异步爬虫或WebSocket demo" }, { "topic": "数据库设计与ORM", "reason": "JD要求熟悉MySQL和Redis,简历仅提到\"了解数据库\"", "tip": "复习SQLAlchemy/FastAPI集成、索引优化、事务隔离级别,准备缓存穿透/雪崩的解决方案" }, { "topic": "系统设计", "reason": "岗位为校招,可能会考察基础设计能力", "tip": "准备短URL服务、秒杀系统等常见设计题,重点练习Python伪代码表达能力" }, ], "encouragement": "你的基础扎实,Java和C++的背景说明你有良好的编程思维。建议集中2-3周时间做一个完整的Python后端项目(如简易电商API),就能在简历上大幅提升匹配度。你的学习能力完全胜任这个方向,加油!" } class Advisor(BaseAgent): name = "advisor" def build_prompt(self, match_report: dict, jd_text: str) -> tuple[str, dict]: system = ( "你是一个求职建议专家。你的任务是:\n" "1. 基于匹配报告,给出精准的简历修改建议。\n" "2. 建议必须包含:哪个维度、具体问题、原文、改写建议、原因。\n" "3. 每条建议必须做到\"HR一眼能看出改变\"的程度,不要泛泛而谈。\n" "4. 如果匹配度低于50分,必须在鼓励中包含'从匹配度角度不推荐'的提示。\n" "5. 提供不超过3个面试准备话题(每个话题包含 topic、reason、tip)。\n" "6. 最后输出一段鼓励语,要求真诚具体,不敷衍。\n\n" "请严格按以下JSON格式输出,字段名必须完全匹配:\n" '{\n' ' "rewrite_suggestions": [\n' ' {"dimension": "技能匹配", "issue": "...", "original": "...", "suggestion": "...", "reason": "..."}\n' ' ],\n' ' "interview_tips": [\n' ' {"topic": "...", "reason": "...", "tip": "..."}\n' ' ],\n' ' "encouragement": "..."\n' '}\n\n' "输出格式为JSON,不要加markdown代码块标记。" ) user = { "match_report": match_report, "jd_text": jd_text } return system, user async def generate_suggestions(self, match_report: dict, jd_text: str) -> dict: api_key = os.environ.get("DEEPSEEK_API_KEY") if not api_key or os.environ.get("MOCK_MODE", "").lower() == "true": return MOCK_RESPONSE system, user_input = self.build_prompt(match_report, jd_text) result = await self.call_llm(system, user_input) raw_suggestions = result.get("rewrite_suggestions") or result.get("suggestions") or result.get("modifications") or [] normalized = [] for s in raw_suggestions: if isinstance(s, dict): normalized.append({ "dimension": s.get("dimension") or s.get("name") or "", "issue": s.get("issue") or s.get("problem") or "", "original": s.get("original") or s.get("original_text") or "", "suggestion": s.get("suggestion") or s.get("rewrite") or "", "reason": s.get("reason") or s.get("rationale") or "", }) result["rewrite_suggestions"] = normalized raw_tips = result.get("interview_tips") or result.get("tips") or result.get("preparation") or [] normalized_tips = [] for t in raw_tips: if isinstance(t, dict): normalized_tips.append({ "topic": t.get("topic") or t.get("title") or "", "reason": t.get("reason") or t.get("why") or "", "tip": t.get("tip") or t.get("advice") or "", }) result["interview_tips"] = normalized_tips if not result.get("encouragement"): result["encouragement"] = "" if "error" in result: return MOCK_RESPONSE return result