diff --git a/build_exe.bat b/build_exe.bat index 24b4796..8bf6fe1 100644 --- a/build_exe.bat +++ b/build_exe.bat @@ -7,7 +7,7 @@ REM 每次升级改 APP_VERSION,避免与正在运行的旧版本目录冲突 REM 使用:把整个文件夹拷到目标机器(或压缩成 zip 发送), REM 双击文件夹里的 Boss直聘AI投递-%APP_VERSION%.exe 运行。 REM ============================================================ -set APP_VERSION=1.10 +set APP_VERSION=1.11 echo [1/2] 安装打包依赖... pip install pyinstaller PySide6 --quiet diff --git a/client/.env.example b/client/.env.example index b0d81cf..302f74d 100644 --- a/client/.env.example +++ b/client/.env.example @@ -17,9 +17,6 @@ BROWSER_MODE=chrome ADSPOWER_API_URL=http://local.adspower.net:50325 ADSPOWER_USER_ID= -# AI 大模型(OpenAI 兼容接口,代码已内置生产配置,仅在需要覆盖时填写) -#AI_BASE_URL=https://api.siliconflow.cn/v1 -#AI_MODEL=deepseek-ai/DeepSeek-V4-Flash -#AI_API_KEY= +# AI 打分经云端服务端代理,客户端无需配置模型密钥 # 低于该分数的职位自动跳过 AI_MIN_SCORE=65 diff --git a/client/ai.py b/client/ai.py index 2955cb0..fa3e53c 100644 --- a/client/ai.py +++ b/client/ai.py @@ -1,102 +1,124 @@ """ -客户端 AI 大模型封装(OpenAI 兼容接口) +客户端 AI 评分封装(经云端服务端代理) +模型配置(OpenAI 兼容 base_url/api_key/model)与密钥全部由服务端持有, +客户端只负责:发起打分任务 → 轮询结果;不再直连任何大模型。 功能: -1. 职位匹配度评分 -2. 简历内容整理 +1. 职位匹配度评分(异步任务:开始打分 + 轮询结果) +2. 简历内容整理(服务端同步代理) """ -import json -import os -from openai import OpenAI +import time +import uuid + +import requests from loguru import logger -from .config import client_settings +from .server_api import ServerAPI + +# 轮询间隔 / 等待总预算(与服务端 5 分钟悬空判定对齐) +POLL_INTERVAL_S = 3 +POLL_BUDGET_S = 300 + + +def gen_request_key() -> str: + """幂等键:uuid1 = 本机 MAC + 时间戳,.hex 为 32 位十六进制 + (与服务端 ai_score_task.request_key VARCHAR(32) 严格对齐)""" + return uuid.uuid1().hex + + +def _detail_of(e: requests.HTTPError) -> str: + """从 HTTPError 中提取服务端 detail 文案""" + try: + return str(e.response.json().get("detail") or e) + except Exception: + return str(e) class MatchResult: - def __init__(self, score: int, reason: str): + def __init__(self, score: int, reason: str, fatal: bool = False): self.score = score self.reason = reason + # fatal = 模型级失败(调用出错/超时/未配置):客户端必须跳过该岗位; + # 非 fatal(服务端不可达等网络问题):按约定降级默认放行 + self.fatal = fatal def is_worth_applying(self, threshold: int = 60) -> bool: return self.score >= threshold class AIClient: - """OpenAI 兼容大模型客户端""" + """AI 评分客户端(服务端代理模式)""" - def __init__(self, api_key: str | None = None, base_url: str | None = None, - model: str | None = None): - self.api_key = api_key or client_settings.AI_API_KEY or os.environ.get("AI_API_KEY", "") - self.base_url = base_url or client_settings.AI_BASE_URL - self.model = model or client_settings.AI_MODEL - self.client = OpenAI(api_key=self.api_key, base_url=self.base_url) + def __init__(self, server: ServerAPI): + self.server = server def available(self) -> bool: - return bool(self.api_key) + """可用条件:已登录云端账号(打分接口需 JWT 鉴权)。 + 模型是否配置由服务端判定,未配置时发起打分会返回明确错误。""" + return bool(self.server.token) # ------------------------------------------------------------------ - # 职位匹配度评分 + # 职位匹配度评分(异步任务 + 轮询) # ------------------------------------------------------------------ - def score_job(self, jd_text: str, resume_content: str, prompt: str | None = None) -> MatchResult: - """根据 JD 与主简历内容评估匹配度""" + def score_job(self, job_ref: str, jd_text: str, resume_id: int | None, + conditions: dict | None = None, + prompt: str | None = None) -> MatchResult: + """发起打分并等待结果。 + - 服务端明确失败(模型未配置/简历缺失/限额/模型报错/超时)→ fatal,调用方跳岗 + - 网络层不可达 → 降级默认放行(保持既有容错约定) + """ if not self.available(): - return MatchResult(50, "AI 未配置,默认放行") + return MatchResult(50, "AI 未接入(未登录系统账号),默认放行") + if not resume_id: + return MatchResult(50, "无主简历,跳过 AI 评分默认放行") - system_prompt = ( - "你是专业招聘顾问。根据职位描述(JD)和候选人简历,输出 JSON 评估结果," - "不要输出 JSON 以外的内容。\n" - '格式:{"score": <0-100整数>, "reason": "<一句话评价>"}\n' - "评分标准:80-100 高度匹配;60-79 基本匹配;40-59 部分匹配;0-39 不匹配。" - ) - user_prompt = f"【职位描述】\n{jd_text}\n\n【候选人简历】\n{resume_content}" + conds = dict(conditions or {}) if prompt: - user_prompt += f"\n\n【用户额外偏好/引导词】\n{prompt}" + conds["prompt"] = prompt + request_key = gen_request_key() + # 1) 开始打分(幂等键保证网络重试不会重复调用大模型) try: - resp = self.client.chat.completions.create( - model=self.model, - messages=[ - {"role": "system", "content": system_prompt}, - {"role": "user", "content": user_prompt}, - ], - temperature=0.2, - max_tokens=256, - ) - raw = resp.choices[0].message.content.strip() - if raw.startswith("```"): - raw = raw.split("```")[1] - if raw.startswith("json"): - raw = raw[4:] - data = json.loads(raw) - return MatchResult(int(data.get("score", 50)), data.get("reason", "")) + self.server.start_ai_score({ + "request_key": request_key, + "resume_id": resume_id, + "jd_text": jd_text, + "job_ref": job_ref, + "conditions": conds, + }) + except requests.HTTPError as e: + return MatchResult(-1, f"AI 打分不可用:{_detail_of(e)}", fatal=True) except Exception as e: - logger.warning(f"AI 评分失败,默认放行:{e}") - return MatchResult(50, f"AI 异常:{e}") + logger.warning(f"AI 打分发起失败(服务端不可达),默认放行:{e}") + return MatchResult(50, "AI 服务不可达,默认放行") + + # 2) 轮询结果:成功返回分数;失败带原因(调用方跳岗);处理中继续等待 + deadline = time.time() + POLL_BUDGET_S + while time.time() < deadline: + time.sleep(POLL_INTERVAL_S) + try: + r = self.server.get_ai_score_result(request_key) + except Exception as e: + logger.debug(f"AI 结果轮询异常,稍后重试:{e}") + continue + status = r.get("status") + if status == "success": + return MatchResult(int(r.get("score") or 0), r.get("reason") or "") + if status == "failed": + return MatchResult(-1, f"AI 打分失败:{r.get('reason') or '未知'}", + fatal=True) + # pending / running:等待下一次轮询 + logger.warning(f"AI 打分等待超时({POLL_BUDGET_S}s),跳过该岗位:{job_ref}") + return MatchResult(-1, f"AI 打分超时({POLL_BUDGET_S // 60} 分钟)", fatal=True) # ------------------------------------------------------------------ - # 简历内容整理 + # 简历内容整理(服务端同步代理,低频调用) # ------------------------------------------------------------------ def clean_resume(self, raw_text: str) -> str: """用 AI 将抓取的杂乱简历文本整理为清晰的结构化文本""" if not self.available(): return raw_text - - system_prompt = ( - "你是简历整理助手。将下面抓取的杂乱简历文本整理为清晰、结构化的纯文本简历," - "保留所有关键信息(姓名可省略,联系方式可保留),去除广告和无关内容。" - "直接输出整理后的文本,不要加任何解释。" - ) try: - resp = self.client.chat.completions.create( - model=self.model, - messages=[ - {"role": "system", "content": system_prompt}, - {"role": "user", "content": raw_text}, - ], - temperature=0.3, - max_tokens=2048, - ) - return resp.choices[0].message.content.strip() + return self.server.clean_resume_via_server(raw_text) except Exception as e: logger.warning(f"AI 整理简历失败,返回原文:{e}") return raw_text @@ -105,13 +127,4 @@ class AIClient: # 连通测试 # ------------------------------------------------------------------ def ping(self) -> bool: - try: - resp = self.client.chat.completions.create( - model=self.model, - messages=[{"role": "user", "content": "回复1"}], - max_tokens=5, - ) - return True - except Exception as e: - logger.error(f"AI 接口连通失败:{e}") - return False + return self.server.ping() diff --git a/client/browser/applicant.py b/client/browser/applicant.py index bba19dd..b00efc9 100644 --- a/client/browser/applicant.py +++ b/client/browser/applicant.py @@ -158,7 +158,9 @@ class Applicant: self.is_running = False self.session_id: int | None = None - self.primary_resume_content: str = "" + self.primary_resume_id: int | None = None + # 本次投递所用条件快照(随每次打分请求上送服务端) + self._score_conditions: dict = {} self.stats = {"applied": 0, "skipped": 0, "current": ""} # 列表页会话(整个任务期间保持一个标签页) @@ -190,7 +192,7 @@ class Applicant: self.session_id = session["id"] primary = self.server.get_primary_resume() - self.primary_resume_content = primary["content"] if primary else "" + self.primary_resume_id = primary["id"] if primary else None self._stop_event.clear() self.is_running = True @@ -241,6 +243,13 @@ class Applicant: def _run(self, keywords, city, salary_min, salary_max, min_score, delay_range): try: + # 投递条件快照:随每次 AI 打分请求上送服务端(简历内容由服务端按 id 自取) + self._score_conditions = { + "keywords": ",".join(keywords), + "city": city, + "salary_min": salary_min, + "salary_max": salary_max, + } self.manager.ensure_browser_started() # 整个任务只开一个列表标签页,模拟真人持续浏览 self._tab = self.cdp.open_tab(JOBS_PAGE) @@ -463,8 +472,17 @@ class Applicant: card.jd_text = self._fetch_jd(card) ai_score, ai_reason = -1, "" - if self.ai and self.ai.available() and self.primary_resume_content: - result = self.ai.score_job(card.jd_text, self.primary_resume_content, prompt=self.prompt) + if self.ai and self.ai.available() and self.primary_resume_id: + result = self.ai.score_job(card.job_id, card.jd_text, + self.primary_resume_id, + conditions=self._score_conditions, + prompt=self.prompt) + if result.fatal: + # 模型级失败(报错/超时/未配置):跳过该岗位,避免盲投 + logger.warning(f"AI 打分失败,跳过岗位:{card.company} - {card.title}" + f" | {result.reason}") + self._report_skip(card, result.reason[:255]) + return 0 ai_score, ai_reason = result.score, result.reason logger.info(f"AI 评分 {ai_score}(及格线 {min_score}):" f"{card.company} - {card.title} | {ai_reason}") diff --git a/client/config.py b/client/config.py index efb07a3..5ce7530 100644 --- a/client/config.py +++ b/client/config.py @@ -34,11 +34,9 @@ class ClientSettings(BaseSettings): ADSPOWER_USER_ID: str = "" # ------------------------------------------------------- - # AI 大模型(OpenAI 兼容接口,默认内置生产配置,.env 可覆盖) + # AI 打分(模型配置与密钥由云端服务端持有并代理调用, + # 客户端只保留本地投递过滤线) # ------------------------------------------------------- - AI_BASE_URL: str = "https://api.siliconflow.cn/v1" - AI_MODEL: str = "deepseek-ai/DeepSeek-V4-Flash" - AI_API_KEY: str = "sk-wptdswlffnkdorhomdgfifhxxbrsuklsloardmclwxjjehpa" AI_MIN_SCORE: int = 65 # ------------------------------------------------------- diff --git a/client/main.py b/client/main.py index 2f9d3de..9547cfa 100644 --- a/client/main.py +++ b/client/main.py @@ -88,8 +88,7 @@ def main() -> None: logger.warning("浏览器模式:chrome,但未检测到 Chrome/Edge,请安装或在 .env 配置 BROWSER_PATH") elif not client_settings.ADSPOWER_USER_ID: logger.warning("浏览器模式:adspower,尚未配置 Profile ID(ADSPOWER_USER_ID)") - if not client_settings.AI_API_KEY: - logger.warning("尚未配置 AI_API_KEY,AI 评分功能将不可用") + # AI 打分经云端服务端代理(模型配置在服务端 /view 维护),客户端无需本地密钥 host = client_settings.WEB_HOST port = client_settings.WEB_PORT diff --git a/client/server_api.py b/client/server_api.py index dd9b592..65b800a 100644 --- a/client/server_api.py +++ b/client/server_api.py @@ -50,17 +50,18 @@ class ServerAPI: def _headers(self) -> dict: return {"Authorization": f"Bearer {self.token}"} if self.token else {} - def _req(self, method: str, url: str, **kwargs) -> requests.Response: - """云端请求统一入口:网络层异常重试 5 次、每次间隔 1 秒,仍失败才抛出。 + def _req(self, method: str, url: str, retries: int = 5, **kwargs) -> requests.Response: + """云端请求统一入口:网络层异常重试(默认 5 次、每次间隔 1 秒),仍失败才抛出。 仅重试连接类异常(RequestException 发生在请求阶段); - HTTP 状态码错误由调用方 raise_for_status 抛出,不重试。""" + HTTP 状态码错误由调用方 raise_for_status 抛出,不重试。 + AI 代理类长调用可传小 retries,避免重试放大等待。""" last: Exception | None = None - for i in range(5): + for i in range(retries): try: return requests.request(method, url, **kwargs) except requests.RequestException as e: last = e - logger.warning(f"云端请求失败({i + 1}/5),1 秒后重试:" + logger.warning(f"云端请求失败({i + 1}/{retries}),1 秒后重试:" f"{method.upper()} {url}:{e}") time.sleep(1) raise last @@ -313,6 +314,33 @@ class ServerAPI: resp.raise_for_status() return resp.json() + # ------------------------------------------------------------------ + # AI 代理(模型配置在服务端,客户端不持有任何密钥) + # ------------------------------------------------------------------ + def start_ai_score(self, payload: dict) -> dict: + """开始打分任务(异步)。payload 含 request_key 幂等键, + 网络重试不会导致服务端重复调用大模型""" + resp = self._req("post", f"{self.base_url}/api/ai/score/start", + json=payload, headers=self._headers(), + timeout=30, retries=2) + resp.raise_for_status() + return resp.json() + + def get_ai_score_result(self, request_key: str) -> dict: + """轮询打分结果:{status: pending/running/success/failed, score, reason}""" + resp = self._req("get", f"{self.base_url}/api/ai/score/result/{request_key}", + headers=self._headers(), timeout=15, retries=2) + resp.raise_for_status() + return resp.json() + + def clean_resume_via_server(self, raw_text: str) -> str: + """经服务端 AI 代理整理简历内容(低频同步,超时对齐服务端 5 分钟阶梯)""" + resp = self._req("post", f"{self.base_url}/api/ai/clean-resume", + json={"raw_text": raw_text}, headers=self._headers(), + timeout=330, retries=1) + resp.raise_for_status() + return resp.json().get("content") or raw_text + def today_applied_count(self) -> int: """当天(北京时间)实际投递数量""" resp = self._req("get", f"{self.base_url}/api/jobs/today-count", headers=self._headers(), timeout=15) diff --git a/client/web/app.py b/client/web/app.py index a1303aa..ba6ff0b 100644 --- a/client/web/app.py +++ b/client/web/app.py @@ -37,7 +37,7 @@ class AppState: def __init__(self): self.server_api = ServerAPI() self.browser = BrowserManager() - self.ai = AIClient() + self.ai = AIClient(self.server_api) self.applicant: Applicant | None = None self.activator: Activator | None = None self.connected = False # 是否已注册/登录云端 diff --git a/docs/schema.sql b/docs/schema.sql index e144fc9..5d5171a 100644 --- a/docs/schema.sql +++ b/docs/schema.sql @@ -181,6 +181,47 @@ CREATE TABLE IF NOT EXISTS rpa_task ( ) ENGINE=InnoDB DEFAULT CHARSET=utf8mb4 COLLATE=utf8mb4_unicode_ci COMMENT='远程控制任务'; +-- ------------------------------------------------------------ +-- 8. AI 模型配置(全局单行,id=1,/view 管理页在线维护) +-- ------------------------------------------------------------ +CREATE TABLE IF NOT EXISTS ai_config ( + id INT NOT NULL DEFAULT 1 PRIMARY KEY, + base_url VARCHAR(255) NULL COMMENT 'OpenAI兼容接口地址,如 https://api.siliconflow.cn/v1', + api_key VARCHAR(255) NULL COMMENT '模型 API Key', + model VARCHAR(128) NULL COMMENT '模型名,如 deepseek-ai/DeepSeek-V3', + timeout_s INT NOT NULL DEFAULT 300 COMMENT '服务端调用大模型超时(秒)', + enabled TINYINT(1) NOT NULL DEFAULT 1 COMMENT '是否启用AI代理', + updated_at DATETIME NULL DEFAULT CURRENT_TIMESTAMP ON UPDATE CURRENT_TIMESTAMP +) ENGINE=InnoDB DEFAULT CHARSET=utf8mb4 COLLATE=utf8mb4_unicode_ci + COMMENT='AI模型配置(全局单行)'; + +-- ------------------------------------------------------------ +-- 9. AI 打分任务(客户端异步评分:开始打分 + 轮询结果) +-- request_key 为客户端 uuid1().hex(MAC+时间戳,32 位),故 VARCHAR(32) +-- ------------------------------------------------------------ +CREATE TABLE IF NOT EXISTS ai_score_task ( + id INT AUTO_INCREMENT PRIMARY KEY, + user_id INT NOT NULL, + request_key VARCHAR(32) NOT NULL COMMENT '幂等键:客户端uuid1().hex(MAC+时间戳,32位)', + resume_id INT NULL COMMENT '主简历ID(服务端从库中取内容)', + job_ref VARCHAR(64) NULL COMMENT 'Boss职位ID(审计追踪用)', + jd_text TEXT NOT NULL COMMENT '职位JD全文(评分输入)', + conditions TEXT NULL COMMENT '投递条件JSON:keywords/city/salary_min/salary_max/prompt', + status VARCHAR(20) NOT NULL DEFAULT 'pending' COMMENT 'pending/running/success/failed', + score INT NULL COMMENT '成功时的匹配分(0-100)', + reason VARCHAR(512) NULL COMMENT '成功时一句话评价;失败时失败原因', + error TEXT NULL COMMENT '详细错误信息(调试用)', + created_at DATETIME NOT NULL DEFAULT CURRENT_TIMESTAMP, + finished_at DATETIME NULL, + UNIQUE KEY uq_user_request_key (user_id, request_key), + KEY idx_user_created (user_id, created_at), + CONSTRAINT fk_ai_score_task_user FOREIGN KEY (user_id) + REFERENCES users (id) ON DELETE CASCADE, + CONSTRAINT fk_ai_score_task_resume FOREIGN KEY (resume_id) + REFERENCES resumes (id) ON DELETE SET NULL +) ENGINE=InnoDB DEFAULT CHARSET=utf8mb4 COLLATE=utf8mb4_unicode_ci + COMMENT='AI打分任务(客户端异步评分)'; + -- -- ============================================================ -- -- 附:老库升级脚本(已建过表、缺少新增字段时执行) -- -- ============================================================ @@ -200,6 +241,9 @@ CREATE TABLE IF NOT EXISTS rpa_task ( -- REFERENCES users (id) ON DELETE CASCADE -- ) ENGINE=InnoDB DEFAULT CHARSET=utf8mb4 COLLATE=utf8mb4_unicode_ci; +-- v1.11 老库升级(AI 服务端代理:ai_config + ai_score_task 两表), +-- 完整可重复执行版本见 docs/upgrade_1.11.sql。 + -- 老库城市字段改名(城市代码 → 城市名称): ALTER TABLE apply_sessions CHANGE COLUMN city_code city VARCHAR(20) NULL COMMENT '城市名称'; diff --git a/server/main.py b/server/main.py index 893dc45..9a570bf 100644 --- a/server/main.py +++ b/server/main.py @@ -14,7 +14,7 @@ from loguru import logger from .config import settings from .database import engine -from .routers import register, resumes, sessions, jobs, blacklist, rpa, templates +from .routers import register, resumes, sessions, jobs, blacklist, rpa, templates, ai @asynccontextmanager @@ -51,6 +51,7 @@ app.include_router(jobs.router) app.include_router(blacklist.router) app.include_router(rpa.router) app.include_router(templates.router) +app.include_router(ai.router) @app.get("/", tags=["健康检查"]) diff --git a/server/models.py b/server/models.py index 0f666e2..504124f 100644 --- a/server/models.py +++ b/server/models.py @@ -242,3 +242,59 @@ class ApplyTemplate(Base): updated_at = Column(DateTime, default=utcnow, onupdate=utcnow, nullable=False) user = relationship("User", back_populates="templates") + + +class AIConfig(Base): + """ + AI 模型配置(全局单行,id=1) + OpenAI 兼容接口,由管理页面(/view)在线维护, + 客户端不再持有任何模型密钥。 + """ + __tablename__ = "ai_config" + + id = Column(Integer, primary_key=True, default=1) + # OpenAI 兼容接口地址,如 https://api.siliconflow.cn/v1 + base_url = Column(String(255), nullable=True) + api_key = Column(String(255), nullable=True) + model = Column(String(128), nullable=True) + # 服务端调用大模型的超时(秒) + timeout_s = Column(Integer, default=300, nullable=False) + # 是否启用 AI 代理(关闭后客户端评分请求直接返回未配置) + enabled = Column(Boolean, default=True, nullable=False) + updated_at = Column(DateTime, default=utcnow, onupdate=utcnow) + + +class AIScoreTask(Base): + """ + AI 打分任务(客户端异步评分) + 客户端发起打分 → 服务端后台调用大模型 → 结果落表 → 客户端轮询取结果。 + request_key 为客户端生成的幂等键(uuid1 hex,MAC+时间戳,32 位), + 重复提交同键直接返回已有任务,不会重复调用大模型。 + """ + __tablename__ = "ai_score_task" + __table_args__ = ( + # 同一用户内幂等键唯一(防重复调用大模型) + UniqueConstraint("user_id", "request_key", name="uq_user_request_key"), + Index("idx_user_created", "user_id", "created_at"), + ) + + id = Column(Integer, primary_key=True, autoincrement=True) + user_id = Column(Integer, ForeignKey("users.id", ondelete="CASCADE"), nullable=False) + # 幂等键:客户端 uuid1().hex,32 位十六进制字符串,故 VARCHAR(32) + request_key = Column(String(32), nullable=False) + resume_id = Column(Integer, ForeignKey("resumes.id", ondelete="SET NULL"), nullable=True) + # 客户端职位标识(Boss job_id,仅审计追踪用) + job_ref = Column(String(64), nullable=True) + # 职位 JD 全文(评分输入) + jd_text = Column(Text, nullable=False) + # 投递条件 JSON:keywords/city/salary_min/salary_max/prompt + conditions = Column(Text, nullable=True) + # pending / running / success / failed + status = Column(String(20), default="pending", nullable=False) + score = Column(Integer, nullable=True) + # 成功时一句话评价;失败时失败原因(客户端据此跳岗) + reason = Column(String(512), nullable=True) + # 详细错误信息(调试用) + error = Column(Text, nullable=True) + created_at = Column(DateTime, default=utcnow, nullable=False) + finished_at = Column(DateTime, nullable=True) diff --git a/server/requirements.txt b/server/requirements.txt index 37768ad..a29a5a1 100644 --- a/server/requirements.txt +++ b/server/requirements.txt @@ -8,3 +8,4 @@ pydantic>=2.7.0 pydantic-settings>=2.3.0 python-jose[cryptography]>=3.3.0 loguru>=0.7.2 +requests>=2.31.0 diff --git a/server/routers/ai.py b/server/routers/ai.py new file mode 100644 index 0000000..bd537cc --- /dev/null +++ b/server/routers/ai.py @@ -0,0 +1,332 @@ +""" +AI 代理路由:模型配置管理 + 异步打分任务 + 简历整理 +- 客户端发起打分(POST /api/ai/score/start)→ 服务端后台线程调用大模型 → 结果落表 +- 客户端轮询结果(GET /api/ai/score/result/{request_key}): + 成功返回 score+reason;失败返回 reason(客户端跳过该岗位);处理中则继续等待 +- request_key 为客户端生成的幂等键(本机 MAC+时间戳的 uuid1 hex,32 位), + 重复提交同键直接返回已有任务,绝不重复调用大模型 +- 模型配置(OpenAI 兼容)由 /view 管理页在线维护,客户端不持有任何密钥 +""" +import json +import threading +from datetime import datetime, timedelta, timezone + +import requests +from fastapi import APIRouter, Depends, HTTPException +from loguru import logger +from sqlalchemy.orm import Session + +from ..auth import get_current_user +from ..database import get_db, SessionLocal +from ..models import AIConfig, AIScoreTask, Resume, User, utcnow +from ..schemas import AIConfigIn, CleanResumeIn, ScoreStartIn + +router = APIRouter(prefix="/api/ai", tags=["AI代理"]) + +# 简易限额:每用户每日允许发起的打分次数(内存计数,重启清零) +DAILY_QUOTA = 2000 +# 处理中任务超过该时长仍无结果 → 判定失败(防悬空等待,5 分钟) +TASK_TIMEOUT_S = 300 + +_quota_lock = threading.Lock() +_quota: dict = {} # {(user_id, 北京日期): 已用次数} + +_SCORE_SYSTEM_PROMPT = ( + "你是专业招聘顾问。根据职位描述(JD)和候选人简历,输出 JSON 评估结果," + "不要输出 JSON 以外的内容。\n" + '格式:{"score": <0-100整数>, "reason": "<一句话评价>"}\n' + "评分标准:80-100 高度匹配;60-79 基本匹配;40-59 部分匹配;0-39 不匹配。" +) +_CLEAN_SYSTEM_PROMPT = ( + "你是简历整理助手。将下面抓取的杂乱简历文本整理为清晰、结构化的纯文本简历," + "保留所有关键信息(姓名可省略,联系方式可保留),去除广告和无关内容。" + "直接输出整理后的文本,不要加任何解释。" +) + + +# ------------------------------------------------------------------ +# 工具函数 +# ------------------------------------------------------------------ +def _today_bj() -> str: + """北京日期(限额按天清零)""" + return datetime.now(timezone(timedelta(hours=8))).strftime("%Y-%m-%d") + + +def _quota_ok(user_id: int) -> bool: + with _quota_lock: + return _quota.get((user_id, _today_bj()), 0) < DAILY_QUOTA + + +def _quota_inc(user_id: int) -> None: + with _quota_lock: + key = (user_id, _today_bj()) + _quota[key] = _quota.get(key, 0) + 1 + + +def _get_config(db: Session) -> AIConfig | None: + return db.query(AIConfig).filter(AIConfig.id == 1).first() + + +def _llm_chat(cfg: AIConfig, system: str, user_text: str, + max_tokens: int, temperature: float) -> str: + """直接 HTTP 调用 OpenAI 兼容 chat/completions(服务端不引入 openai SDK)""" + url = (cfg.base_url or "").rstrip("/") + "/chat/completions" + messages = [] + if system: + messages.append({"role": "system", "content": system}) + messages.append({"role": "user", "content": user_text}) + resp = requests.post( + url, + headers={"Authorization": f"Bearer {cfg.api_key}"}, + json={"model": cfg.model, "messages": messages, + "temperature": temperature, "max_tokens": max_tokens}, + timeout=cfg.timeout_s or 300, + ) + resp.raise_for_status() + data = resp.json() + return (data["choices"][0]["message"]["content"] or "").strip() + + +def parse_score(raw: str) -> tuple[int, str]: + """解析模型返回的评分 JSON(纯函数,便于单测)""" + raw = (raw or "").strip() + if raw.startswith("```"): + raw = raw.split("```")[1] + if raw.startswith("json"): + raw = raw[4:] + data = json.loads(raw) + return int(data.get("score", 50)), str(data.get("reason", ""))[:500] + + +def build_score_user_prompt(jd_text: str, resume_content: str, + conditions: dict | None) -> str: + """拼装评分用户提示词(纯函数,便于单测)""" + user_prompt = f"【职位描述】\n{jd_text}\n\n【候选人简历】\n{resume_content}" + conds = [] + if conditions: + if conditions.get("keywords"): + conds.append(f"搜索关键词:{conditions['keywords']}") + if conditions.get("city"): + conds.append(f"目标城市:{conditions['city']}") + if conditions.get("salary_min") or conditions.get("salary_max"): + conds.append(f"期望薪资:{conditions.get('salary_min') or '?'}" + f"-{conditions.get('salary_max') or '?'}K") + if conds: + user_prompt += "\n\n【本次投递条件】\n" + "\n".join(conds) + if conditions and conditions.get("prompt"): + user_prompt += f"\n\n【用户额外偏好/引导词】\n{conditions['prompt']}" + return user_prompt + + +# ------------------------------------------------------------------ +# 异步打分:开始 + 查询结果 +# ------------------------------------------------------------------ +@router.post("/score/start") +def score_start( + req: ScoreStartIn, + current_user: User = Depends(get_current_user), + db: Session = Depends(get_db), +): + """开始打分任务(幂等:同 request_key 重复提交直接返回已有任务)""" + task = db.query(AIScoreTask).filter( + AIScoreTask.user_id == current_user.id, + AIScoreTask.request_key == req.request_key, + ).first() + if task: + logger.debug(f"幂等命中,直接返回已有打分任务:{req.request_key}") + return {"request_key": task.request_key, "status": task.status} + + cfg = _get_config(db) + if not cfg or not cfg.enabled or not cfg.api_key: + raise HTTPException(status_code=400, detail="服务端未配置 AI 模型,请联系管理员") + resume = db.query(Resume).filter( + Resume.id == req.resume_id, + Resume.user_id == current_user.id, + ).first() + if not resume or not (resume.content or "").strip(): + raise HTTPException(status_code=404, detail="简历不存在或内容为空") + if not _quota_ok(current_user.id): + raise HTTPException(status_code=429, detail="今日 AI 打分次数已用完") + + _quota_inc(current_user.id) + conditions = req.conditions.model_dump() if req.conditions else {} + task = AIScoreTask( + user_id=current_user.id, + request_key=req.request_key, + resume_id=resume.id, + job_ref=req.job_ref, + jd_text=req.jd_text, + conditions=json.dumps(conditions, ensure_ascii=False), + status="pending", + ) + db.add(task) + db.commit() + db.refresh(task) + + threading.Thread(target=_run_score_task, args=(task.id,), daemon=True).start() + logger.info(f"AI 打分任务已受理:task={task.id} user={current_user.id} " + f"job_ref={req.job_ref}") + return {"request_key": task.request_key, "status": task.status} + + +def _run_score_task(task_id: int) -> None: + """后台执行打分:独立 DB 会话,成功/失败结果一律落表""" + db = SessionLocal() + try: + task = db.query(AIScoreTask).filter(AIScoreTask.id == task_id).first() + if not task: + return + task.status = "running" + db.commit() + + cfg = _get_config(db) + if not cfg or not cfg.api_key: + raise RuntimeError("AI 模型配置缺失或已停用") + resume = (db.query(Resume).filter(Resume.id == task.resume_id).first() + if task.resume_id else None) + resume_content = ((resume.content if resume else "") or "") + + try: + conditions = json.loads(task.conditions or "{}") + except Exception: + conditions = {} + + user_prompt = build_score_user_prompt(task.jd_text, resume_content, conditions) + raw = _llm_chat(cfg, _SCORE_SYSTEM_PROMPT, user_prompt, + max_tokens=256, temperature=0.2) + score, reason = parse_score(raw) + + task.status = "success" + task.score = score + task.reason = reason + task.finished_at = utcnow() + db.commit() + logger.success(f"AI 打分完成:task={task_id} score={score}") + except Exception as e: + logger.error(f"AI 打分任务失败 task={task_id}:{e}") + try: + db.rollback() + task = db.query(AIScoreTask).filter(AIScoreTask.id == task_id).first() + if task and task.status in ("pending", "running"): + task.status = "failed" + task.reason = f"模型调用失败:{str(e)[:200]}" + task.error = str(e)[:2000] + task.finished_at = utcnow() + db.commit() + except Exception: + db.rollback() + finally: + db.close() + + +@router.get("/score/result/{request_key}") +def score_result( + request_key: str, + current_user: User = Depends(get_current_user), + db: Session = Depends(get_db), +): + """查询打分结果(客户端轮询) + status:pending/running=处理中(客户端继续等待下一次轮询); + success=返回 score+reason;failed=返回 reason(客户端跳过该岗位)。 + 处理中的任务若超 5 分钟仍无结果,判定为失败(防悬空)。 + """ + task = db.query(AIScoreTask).filter( + AIScoreTask.user_id == current_user.id, + AIScoreTask.request_key == request_key, + ).first() + if not task: + raise HTTPException(status_code=404, detail="打分任务不存在") + if task.status in ("pending", "running") and \ + (utcnow() - task.created_at).total_seconds() > TASK_TIMEOUT_S: + task.status = "failed" + task.reason = "打分超时(超过 5 分钟无结果)" + task.finished_at = utcnow() + db.commit() + return {"status": task.status, "score": task.score, "reason": task.reason} + + +# ------------------------------------------------------------------ +# 简历整理(低频同步调用,简历抓取时用) +# ------------------------------------------------------------------ +@router.post("/clean-resume") +def clean_resume( + req: CleanResumeIn, + current_user: User = Depends(get_current_user), + db: Session = Depends(get_db), +): + cfg = _get_config(db) + if not cfg or not cfg.enabled or not cfg.api_key: + raise HTTPException(status_code=400, detail="服务端未配置 AI 模型") + try: + text = _llm_chat(cfg, _CLEAN_SYSTEM_PROMPT, req.raw_text, + max_tokens=2048, temperature=0.3) + except Exception as e: + raise HTTPException(status_code=502, detail=f"AI 整理失败:{str(e)[:200]}") + return {"content": text} + + +# ------------------------------------------------------------------ +# 模型配置(管理页面在线维护) +# ------------------------------------------------------------------ +def _config_out(cfg: AIConfig, with_key: bool) -> dict: + return { + "base_url": cfg.base_url or "", + "api_key": cfg.api_key if with_key else "", + "model": cfg.model or "", + "timeout_s": cfg.timeout_s, + "enabled": cfg.enabled, + "updated_at": cfg.updated_at, + } + + +@router.get("/config") +def get_ai_config( + with_key: bool = False, + current_user: User = Depends(get_current_user), + db: Session = Depends(get_db), +): + """获取模型配置(默认不回显密钥;编辑时带 with_key=true 回显)""" + cfg = _get_config(db) + if not cfg: + return {"base_url": "", "api_key": "", "model": "", + "timeout_s": 300, "enabled": False, "updated_at": None} + return _config_out(cfg, with_key=with_key) + + +@router.put("/config") +def save_ai_config( + req: AIConfigIn, + current_user: User = Depends(get_current_user), + db: Session = Depends(get_db), +): + """保存模型配置(OpenAI 兼容接口)""" + cfg = _get_config(db) + if not cfg: + cfg = AIConfig(id=1) + db.add(cfg) + cfg.base_url = (req.base_url or "").strip() + cfg.api_key = (req.api_key or "").strip() + cfg.model = (req.model or "").strip() + cfg.timeout_s = req.timeout_s + cfg.enabled = req.enabled + db.commit() + db.refresh(cfg) + logger.info(f"AI 模型配置已更新(by user {current_user.id}):" + f"{cfg.base_url} / {cfg.model} / enabled={cfg.enabled}") + return _config_out(cfg, with_key=False) + + +@router.post("/config/test") +def test_ai_config( + current_user: User = Depends(get_current_user), + db: Session = Depends(get_db), +): + """连通测试:用当前已保存配置发起一次最小请求""" + cfg = _get_config(db) + if not cfg or not cfg.api_key: + return {"ok": False, "detail": "尚未配置 API Key"} + try: + _llm_chat(cfg, "", "回复1", max_tokens=5, temperature=0) + return {"ok": True, "detail": "连通正常"} + except Exception as e: + return {"ok": False, "detail": str(e)[:300]} diff --git a/server/schemas.py b/server/schemas.py index f799c2d..4b9298e 100644 --- a/server/schemas.py +++ b/server/schemas.py @@ -239,6 +239,43 @@ class Message(BaseModel): msg: str +# ======================================================= +# AI 代理(异步打分 + 模型配置) +# ======================================================= +class ScoreConditions(BaseModel): + """本次投递所用条件(客户端每次随打分请求上送)""" + keywords: Optional[str] = None + city: Optional[str] = None + salary_min: Optional[int] = None + salary_max: Optional[int] = None + # AI 评分引导词(用户自定义额外偏好) + prompt: Optional[str] = None + + +class ScoreStartIn(BaseModel): + # 幂等键:客户端用本机 MAC+时间戳生成的 uuid1 hex(32 位十六进制) + request_key: str = Field(..., min_length=32, max_length=32, + pattern=r"^[0-9a-fA-F]{32}$", + description="uuid1().hex,32 位十六进制") + resume_id: int = Field(..., description="主简历 ID,服务端从库中取内容") + jd_text: str = Field(..., min_length=1, description="职位 JD 全文") + job_ref: Optional[str] = Field(None, max_length=64, description="Boss 职位 ID(审计用)") + conditions: Optional[ScoreConditions] = None + + +class CleanResumeIn(BaseModel): + raw_text: str = Field(..., min_length=1, description="抓取到的原始简历文本") + + +class AIConfigIn(BaseModel): + """模型配置(OpenAI 兼容接口,/view 管理页维护)""" + base_url: str = Field("", max_length=255, description="如 https://api.siliconflow.cn/v1") + api_key: str = Field("", max_length=255) + model: str = Field("", max_length=128) + timeout_s: int = Field(300, ge=10, le=600, description="服务端调用大模型超时(秒)") + enabled: bool = True + + class StatsOut(BaseModel): total_applied: int total_chat: int diff --git a/server/static/view.html b/server/static/view.html index 500743b..914137a 100644 --- a/server/static/view.html +++ b/server/static/view.html @@ -112,6 +112,7 @@
+ @@ -244,6 +245,25 @@ + + +