AI提到后端

This commit is contained in:
2026-08-26 20:13:34 +08:00
parent fae7a51cc5
commit 80bf1d218c
15 changed files with 678 additions and 96 deletions
+1 -4
View File
@@ -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
+86 -73
View File
@@ -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()
+22 -4
View File
@@ -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}")
+2 -4
View File
@@ -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
# -------------------------------------------------------
+1 -2
View File
@@ -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
+33 -5
View File
@@ -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)
+1 -1
View File
@@ -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