feat: 简历解析优先用大模型提取字段,失败回退正则

backend/app/routers/ingest.py

@@ -18,7 +18,7 @@ from backend.app.repositories.ingest_repository import (
 )
 from backend.app.security import verify_ingest_token
 from backend.app.services.file_storage import save_resume_bytes
-from backend.app.services.resume_parser import parse_resume_path
+from backend.app.services.resume_parser import parse_resume_path_async
 
 router = APIRouter()
 
@@ -50,7 +50,7 @@ async def ingest(request: Request, db: Session = Depends(get_db)):
             content = await download_pdf(pdf_url, raw.get("cookies") or {})
             saved_file = save_resume_bytes(content, "platform-resume.pdf", "ingest")
             try:
-                parsed = parse_resume_path(Path(saved_file["storagePath"]))
+                parsed = await parse_resume_path_async(Path(saved_file["storagePath"]))
             except Exception as exc:
                 parsed = {"parseWarning": str(exc), "resumeText": ""}
         except Exception as exc:

backend/app/routers/resume.py

@@ -2,7 +2,7 @@ from fastapi import APIRouter, File, HTTPException, UploadFile
 from fastapi.responses import FileResponse
 
 from backend.app.services.file_storage import resolve_resume_path, save_resume_bytes
-from backend.app.services.resume_parser import parse_resume_path, parse_uploaded_bytes
+from backend.app.services.resume_parser import parse_resume_path_async, parse_uploaded_bytes_async
 
 router = APIRouter()
 
@@ -11,7 +11,7 @@ router = APIRouter()
 async def parse_resume_upload(resume: UploadFile = File(...)) -> dict:
     content = await resume.read()
     try:
-        return parse_uploaded_bytes(content, resume.filename or "resume.bin")
+        return await parse_uploaded_bytes_async(content, resume.filename or "resume.bin")
     except Exception as exc:
         raise HTTPException(status_code=500, detail=str(exc)) from exc
 
@@ -21,7 +21,7 @@ async def upload_resume(resume: UploadFile = File(...)) -> dict:
     content = await resume.read()
     saved = save_resume_bytes(content, resume.filename or "resume.bin", "upload")
     try:
-        parsed = parse_resume_path(resolve_resume_path(saved["storedName"]))
+        parsed = await parse_resume_path_async(resolve_resume_path(saved["storedName"]))
     except Exception as exc:
         parsed = {"parseWarning": str(exc)}
     return {"ok": True, "file": {**saved, "url": f"/api/resumes/{saved['storedName']}"}, "parsed": parsed}

backend/app/services/llm.py

@@ -217,6 +217,71 @@ def sanitize_candidate_for_llm(candidate: dict[str, Any]) -> dict[str, Any]:
     }
 
 
+# 简历结构化提取：LLM 返回的字段集合（与 llm_resume_fields 的 prompt 保持一致）。
+# 其中姓名/手机/邮箱/出生年月/年龄等 PII 会被本地脱敏，LLM 无法看到真实值，
+# 这些字段最终以正则解析结果为准（见 resume_parser._merge_parse_result）。
+LLM_RESUME_FIELDS = (
+    "name",
+    "jobTitle",
+    "phone",
+    "email",
+    "birthDate",
+    "age",
+    "years",
+    "education",
+    "school",
+    "major",
+    "city",
+    "skills",
+)
+
+
+async def llm_resume_fields(text: str, hints: dict[str, Any] | None = None) -> dict[str, Any] | None:
+    """用大模型从简历文本中提取结构化字段；未配置 Key 或调用失败时返回 None（调用方回退正则）。
+
+    发送前用正则解析结果对姓名/手机/邮箱/城市/出生年月/年龄做本地脱敏，
+    被脱敏的字段 LLM 一律返回空值，最终以正则结果为准。
+    """
+    settings = get_settings()
+    if not settings.effective_llm_api_key:
+        return None
+    sanitized = sanitize_resume_for_llm(text, hints or {})
+    if not sanitized.strip():
+        return None
+    user = (
+        "请从下面的简历文本中提取结构化字段，只返回一个 JSON 对象（不要 Markdown 围栏，不要多余解释）。\n\n"
+        "字段与规则：\n"
+        '- name: 姓名（2-4 个中文字符）。姓名已被 [本地脱敏] 占位符替换时返回空字符串 ""。\n'
+        '- jobTitle: 求职意向/应聘岗位，只能是简历中明确写出的岗位名（如"销售支持专员"）；没有明确写出时返回 ""，'
+        "不要根据教育/技能内容推断岗位。\n"
+        '- phone: 手机号。已被 [手机号已本地脱敏] 替换或没有时返回 ""。\n'
+        '- email: 邮箱。已被 [邮箱已本地脱敏] 替换或没有时返回 ""。\n'
+        '- birthDate: 出生日期，格式 YYYY-MM-DD。已被 [出生年月已本地脱敏] 替换或没有时返回 ""。\n'
+        '- age: 年龄数字（如 28）。已被 [年龄已本地脱敏] 替换或没有时返回 ""。\n'
+        '- years: 工作年限，如 "5年" / "3.5年" / "应届生"；简历没有明确写工作年限时返回 ""，不要从无关时间推算。\n'
+        "- education: 学历，只能是 博士/硕士/本科/大专/中专/高中/其他 之一；没有明确信息返回 \"\"。\n"
+        '- school: 毕业院校全称（如"甘肃农业大学"）；没有则返回 ""。\n'
+        '- major: 专业名称（如"农林经济管理"）。主修课程列表里的课程名不是专业，应取"专业"字段或紧邻学校/时间行后的'
+        '独立专业名；没有明确专业返回 ""。\n'
+        '- city: 现居城市或简历中的城市名（如"太原"，不含"市"）；没有则返回 ""。\n'
+        "- skills: 技能数组，最多 12 项，只取简历中明确提到的技能。\n\n"
+        "要求：只提取简历中明确出现的信息，绝不编造或猜测；已被脱敏占位符替换的字段一律返回空字符串/空数组。\n\n"
+        f"简历文本：\n{sanitized[:16000]}"
+    )
+    try:
+        parsed = await _chat_json(
+            settings,
+            "你是企业招聘系统中的资深简历结构化解析助手。只返回 JSON，不要输出 Markdown。",
+            user,
+            temperature=0,
+        )
+    except Exception:
+        return None
+    if not isinstance(parsed, dict):
+        return None
+    return {key: parsed.get(key) for key in LLM_RESUME_FIELDS}
+
+
 async def llm_resume_analysis(payload: dict[str, Any]) -> dict[str, Any] | None:
     settings = get_settings()
     if not settings.effective_llm_api_key:

backend/app/services/resume_parser.py

@@ -2,10 +2,12 @@ from pathlib import Path
 from tempfile import TemporaryDirectory
 
 from backend.app.services.file_storage import sanitize_filename
-from backend.app.services.resume_parser_core import parse_resume
+from backend.app.services.llm import LLM_RESUME_FIELDS, llm_resume_fields
+from backend.app.services.resume_parser_core import parse_resume, school_tags
 
 
 def parse_resume_path(path: Path) -> dict:
+    """纯正则解析入口（无 LLM 时的兜底/调试路径）。"""
     return parse_resume(path)
 
 
@@ -14,3 +16,38 @@ def parse_uploaded_bytes(content: bytes, original_name: str = "resume.bin") -> d
         path = Path(temp_dir) / sanitize_filename(original_name)
         path.write_bytes(content)
         return parse_resume_path(path)
+
+
+async def parse_resume_path_async(path: Path) -> dict:
+    """LLM 优先、正则兜底的简历解析入口（API/CLI 均走这里）。"""
+    rule_result = parse_resume(path)
+    llm_result = await llm_resume_fields(rule_result.get("resumeText") or "", rule_result)
+    return _merge_parse_result(rule_result, llm_result)
+
+
+async def parse_uploaded_bytes_async(content: bytes, original_name: str = "resume.bin") -> dict:
+    with TemporaryDirectory(prefix="recruitment-resume-") as temp_dir:
+        path = Path(temp_dir) / sanitize_filename(original_name)
+        path.write_bytes(content)
+        return await parse_resume_path_async(path)
+
+
+def _merge_parse_result(rule_result: dict, llm_result: dict | None) -> dict:
+    """LLM 结果优先、正则结果兜底：LLM 字段非空则覆盖，否则保留正则值。
+
+    技能取并集去重（LLM 优先）；schoolTags 依据最终 school 重新推导。
+    """
+    result = dict(rule_result)
+    if not isinstance(llm_result, dict):
+        return result
+    for key in LLM_RESUME_FIELDS:
+        value = llm_result.get(key)
+        if isinstance(value, str):
+            value = value.strip()
+        if value not in (None, "", [], {}):
+            result[key] = value
+    llm_skills = [str(item).strip() for item in llm_result.get("skills") or [] if str(item).strip()]
+    if llm_skills:
+        result["skills"] = list(dict.fromkeys([*llm_skills, *(rule_result.get("skills") or [])]))
+    result["schoolTags"] = school_tags(result.get("school") or "")
+    return result

backend/scripts/parse_resume.py

 """Command-line entry point for the backend resume parser."""
 
-from backend.app.services.resume_parser_core import main
+import argparse
+import asyncio
+import json
+from pathlib import Path
+
+from backend.app.services.resume_parser import parse_resume_path_async
+
+
+async def _parse(path: Path) -> None:
+    print(json.dumps(await parse_resume_path_async(path), ensure_ascii=False))
+
+
+def main() -> int:
+    parser = argparse.ArgumentParser(description="解析 PDF/DOCX/TXT 简历并输出 JSON。")
+    parser.add_argument("file", help="待解析的简历文件路径")
+    args = parser.parse_args()
+    try:
+        asyncio.run(_parse(Path(args.file)))
+        return 0
+    except Exception as exc:
+        print(json.dumps({"error": str(exc)}, ensure_ascii=False))
+        return 1
+
 
 if __name__ == "__main__":
     raise SystemExit(main())

backend/tests/test_resume_parser_llm.py

+import asyncio
+
+import pytest
+from backend.app.services.llm import llm_resume_fields
+from backend.app.services.resume_parser import _merge_parse_result, parse_resume_path_async
+
+
+@pytest.fixture()
+def llm_disabled(monkeypatch):
+    """隔离测试环境：不加载 .env、不配置任何 LLM Key，确保走正则兜底。"""
+    monkeypatch.setenv("RECRUITMENT_SKIP_ENV_FILES", "1")
+    for key in ("LLM_API_KEY", "DEEPSEEK_API_KEY", "OPENAI_API_KEY"):
+        monkeypatch.setenv(key, "")
+    import backend.app.config as config
+
+    config.get_settings.cache_clear()
+    yield
+    config.get_settings.cache_clear()
+
+
+def test_merge_parse_result_prefers_llm_values():
+    rule = {
+        "name": "张晓燕",
+        "jobTitle": "",
+        "school": "甘肃农业大学",
+        "major": "人力资源管理",
+        "skills": ["市场营销"],
+        "schoolTags": ["非985/211"],
+    }
+    llm = {
+        "name": "",
+        "jobTitle": "销售支持专员",
+        "phone": "",
+        "email": "",
+        "birthDate": "",
+        "age": "",
+        "years": "",
+        "education": "",
+        "school": "",
+        "major": "农林经济管理",
+        "city": "",
+        "skills": ["销售支持", "客户管理"],
+    }
+    merged = _merge_parse_result(rule, llm)
+    assert merged["jobTitle"] == "销售支持专员"
+    assert merged["major"] == "农林经济管理"
+    assert merged["name"] == "张晓燕"
+    assert merged["skills"] == ["销售支持", "客户管理", "市场营销"]
+
+
+def test_merge_parse_result_keeps_rule_when_llm_empty():
+    rule = {
+        "name": "张三",
+        "jobTitle": "销售支持专员",
+        "school": "甘肃农业大学",
+        "major": "农林经济管理",
+        "skills": [],
+        "schoolTags": ["非985/211"],
+    }
+    llm = {
+        "name": "",
+        "jobTitle": "",
+        "phone": "",
+        "email": "",
+        "birthDate": "",
+        "age": "",
+        "years": "",
+        "education": "",
+        "school": "",
+        "major": "",
+        "city": "",
+        "skills": [],
+    }
+    merged = _merge_parse_result(rule, llm)
+    assert merged["name"] == "张三"
+    assert merged["jobTitle"] == "销售支持专员"
+    assert merged["major"] == "农林经济管理"
+    assert merged["skills"] == []
+
+
+def test_merge_parse_result_recomputes_school_tags():
+    merged = _merge_parse_result(
+        {"school": "甘肃农业大学", "schoolTags": ["非985/211"]},
+        {"school": "清华大学"},
+    )
+    assert merged["school"] == "清华大学"
+    assert merged["schoolTags"] == ["985", "211"]
+
+
+def test_llm_resume_fields_returns_none_without_key(llm_disabled):
+    assert asyncio.run(llm_resume_fields("姓名：张三", {})) is None
+
+
+def test_async_parse_falls_back_without_llm(tmp_path, llm_disabled):
+    resume = tmp_path / "李四_高级前端工程师.txt"
+    resume.write_text(
+        "姓名：李四\n电话：13912345678\n邮箱：lisi@example.com\n本科 北京大学 软件工程\n熟悉 React Python",
+        encoding="utf-8",
+    )
+    parsed = asyncio.run(parse_resume_path_async(resume))
+    assert parsed["name"] == "李四"
+    assert parsed["phone"] == "13912345678"
+    assert parsed["school"] == "北京大学"
+    assert parsed["major"] == "软件工程"