Refa: Resume parsing module (architectural optimizations based on SmartResume Pipeline) (#13255)

Core optimizations (refer to arXiv:2510.09722): 1. PDF text fusion: Metadata + OCR dual-path extraction and fusion 2. Page-aware reconstruction: YOLOv10 page segmentation + hierarchical sorting + line number indexing 3. Parallel task decomposition: Basic information/work experience/educational background three-way parallel LLM extraction 4. Index pointer mechanism: LLM returns a range of line numbers instead of generating the full text, reducing the illusion of full text. --------- Co-authored-by: Aron.Yao <yaowei@yaoweideMacBook-Pro.local> Co-authored-by: Aron.Yao <yaowei@192.168.1.68> Co-authored-by: Yingfeng <yingfeng.zhang@gmail.com>
2026-06-29 23:41:12 +08:00 · 2026-03-02 19:05:50 +08:00
parent 7d6f20585f
commit f8c91e8854
11 changed files with 2810 additions and 144 deletions
--- a/rag/app/resume.py
+++ b/rag/app/resume.py
--- a/rag/prompts/resume_basic_info.md
+++ b/rag/prompts/resume_basic_info.md
@@ -0,0 +1,39 @@
+请从以下带行号索引的简历文本中提取基本信息。
+
+{indexed_text}
+
+提取如下信息到 JSON，若某些字段不存在则输出 "" 空或 0:
+{{
+  "name_kwd": "",
+  "gender_kwd": "",
+  "age_int": 0,
+  "phone_kwd": "",
+  "email_tks": "",
+  "birth_dt": "",
+  "work_exp_flt": 0,
+  "current_location": "",
+  "expect_city_names_tks": [],
+  "expect_position_name_tks": [],
+  "skill_tks": [],
+  "language_tks": [],
+  "certificate_tks": [],
+  "self_evaluation_tks": ""
+}}
+
+字段说明:
+- name_kwd: 姓名，如"张三"
+- gender_kwd: 男/女，若不存在则不填
+- age_int: 当前年龄，整数
+- phone_kwd: 电话/手机，请保留原文中的形式，保留国家码区号括号
+- email_tks: 邮箱，如 "xxx@qq.com"
+- birth_dt: 出生年月，如 "1996-11"
+- work_exp_flt: 工作年限，浮点数
+- current_location: 现居地/当前城市，不要从工作经历中推测，要写明现居地
+- expect_city_names_tks: 期望工作城市列表，简历中需要明确说明是期望城市
+- expect_position_name_tks: 期望职位列表
+- skill_tks: 技能/技术栈列表
+- language_tks: 语言能力列表
+- certificate_tks: 证书/资质列表
+- self_evaluation_tks: 自我评价/个人优势/个人总结，完整提取原文内容
+
+只返回 JSON。 /no_think
--- a/rag/prompts/resume_basic_info_en.md
+++ b/rag/prompts/resume_basic_info_en.md
@@ -0,0 +1,39 @@
+Please extract basic information from the following line-indexed resume text.
+
+{indexed_text}
+
+Extract the following information into JSON. If a field does not exist, output "" or 0:
+{{
+  "name_kwd": "",
+  "gender_kwd": "",
+  "age_int": 0,
+  "phone_kwd": "",
+  "email_tks": "",
+  "birth_dt": "",
+  "work_exp_flt": 0,
+  "current_location": "",
+  "expect_city_names_tks": [],
+  "expect_position_name_tks": [],
+  "skill_tks": [],
+  "language_tks": [],
+  "certificate_tks": [],
+  "self_evaluation_tks": ""
+}}
+
+Field descriptions:
+- name_kwd: Full name, e.g. "John Smith"
+- gender_kwd: Male/Female, leave empty if not present
+- age_int: Current age, integer
+- phone_kwd: Phone number, keep original format including country code and brackets
+- email_tks: Email address, e.g. "xxx@gmail.com"
+- birth_dt: Date of birth, e.g. "1996-11"
+- work_exp_flt: Years of work experience, float
+- current_location: Current city/location, do not infer from work experience, must be explicitly stated
+- expect_city_names_tks: List of preferred work cities, must be explicitly stated in the resume
+- expect_position_name_tks: List of desired positions
+- skill_tks: List of skills/tech stack
+- language_tks: List of language proficiencies
+- certificate_tks: List of certificates/qualifications
+- self_evaluation_tks: Self-evaluation/personal strengths/summary, extract full original text
+
+Return JSON only. /no_think
--- a/rag/prompts/resume_education.md
+++ b/rag/prompts/resume_education.md
@@ -0,0 +1,31 @@
+请从以下带行号索引的简历文本中提取教育背景。
+
+{indexed_text}
+
+提取为 JSON:
+{{
+  "education": [
+    {{
+      "school": "",
+      "major": "",
+      "degree": "",
+      "department": "",
+      "start_date": "",
+      "end_date": "",
+      "desc_lines": [start_index, end_index]
+    }}
+  ]
+}}
+
+字段说明:
+- school: 学校全称，如"厦门大学"，中英文都可以
+- major: 专业，如"机械工程"
+- degree: 学位，本科/硕士/博士/专科/高中/初中，若不存在则填""
+- department: 系/学院，如"信息工程系"
+- start_date: 开始时间，格式为 %Y.%m 或 %Y
+- end_date: 结束时间，若至今填写"至今"，若不存在填写""
+- desc_lines: [起始行号, 结束行号]，教育描述对应的行号范围（可选）
+  - 包括课程成绩、研究方向、GPA、荣誉奖项等
+  - 不存在则填 []
+
+只返回 JSON。 /no_think
--- a/rag/prompts/resume_education_en.md
+++ b/rag/prompts/resume_education_en.md
@@ -0,0 +1,31 @@
+Please extract education background from the following line-indexed resume text.
+
+{indexed_text}
+
+Extract into JSON:
+{{
+  "education": [
+    {{
+      "school": "",
+      "major": "",
+      "degree": "",
+      "department": "",
+      "start_date": "",
+      "end_date": "",
+      "desc_lines": [start_index, end_index]
+    }}
+  ]
+}}
+
+Field descriptions:
+- school: Full school name, e.g. "Stanford University", both Chinese and English are acceptable
+- major: Major/field of study, e.g. "Computer Science"
+- degree: Degree level - Bachelor/Master/PhD/Associate/High School/Middle School, leave "" if not available
+- department: Department/College, e.g. "School of Engineering"
+- start_date: Start date, format %Y.%m or %Y
+- end_date: End date, use "Present" if still enrolled, "" if not available
+- desc_lines: [start_line, end_line], line number range for education description (optional)
+  - Includes coursework, research focus, GPA, honors/awards, etc.
+  - Use [] if not available
+
+Return JSON only. /no_think
--- a/rag/prompts/resume_project_exp.md
+++ b/rag/prompts/resume_project_exp.md
@@ -0,0 +1,31 @@
+请从以下带行号索引的简历文本中提取项目经验。
+
+{indexed_text}
+
+提取为 JSON，每段项目经验包含:
+{{
+  "projectExperience": [
+    {{
+      "project_name": "",
+      "role": "",
+      "start_date": "",
+      "end_date": "",
+      "desc_lines": [start_index, end_index]
+    }}
+  ]
+}}
+
+字段说明:
+- project_name: 项目名称
+- role: 担任角色/职责，如"项目负责人"、"后端开发"
+- start_date: 开始时间，格式为 %Y.%m 或 %Y
+- end_date: 结束时间，若至今填写"至今"，若不存在填写""
+- desc_lines: [起始行号, 结束行号]，项目描述对应的行号范围（整数数组）
+  - 指项目描述的原文引用段落 index 范围，包括项目内容、技术栈、成果等
+  - 不包括 project_name、role、start_date、end_date 所在行
+  - 尽可能写全，直到下一段项目经验或其他段落标题为止
+  - 遇到以下段落标题时必须截止，不要将其包含在 desc_lines 中：
+    个人评价、自我评价、个人总结、个人优势、自我描述、技能特长、专业技能、教育背景、教育经历、工作经历、工作经验、证书资质、语言能力、兴趣爱好、求职意向
+  - 如果不存在就写 []
+
+只返回 JSON。 /no_think
--- a/rag/prompts/resume_project_exp_en.md
+++ b/rag/prompts/resume_project_exp_en.md
@@ -0,0 +1,31 @@
+Please extract project experience from the following line-indexed resume text.
+
+{indexed_text}
+
+Extract into JSON, each project experience entry contains:
+{{
+  "projectExperience": [
+    {{
+      "project_name": "",
+      "role": "",
+      "start_date": "",
+      "end_date": "",
+      "desc_lines": [start_index, end_index]
+    }}
+  ]
+}}
+
+Field descriptions:
+- project_name: Project name
+- role: Role/responsibility, e.g. "Project Lead", "Backend Developer"
+- start_date: Start date, format %Y.%m or %Y
+- end_date: End date, use "Present" if ongoing, "" if not available
+- desc_lines: [start_line, end_line], line number range for project description (integer array)
+  - Refers to the original text reference range for project description, including project content, tech stack, achievements, etc.
+  - Does not include lines containing project_name, role, start_date, end_date
+  - Include as much as possible until the next project experience entry or other section heading
+  - STOP before these section headings (do not include them in desc_lines):
+    Self-evaluation, Personal Summary, Skills, Technical Skills, Education, Work Experience, Certificates, Languages, Hobbies, Career Objective
+  - Use [] if not available
+
+Return JSON only. /no_think
--- a/rag/prompts/resume_system.md
+++ b/rag/prompts/resume_system.md
@@ -0,0 +1,3 @@
+你是一个专业的简历分析助手。你的任务是将给定的简历文本转换为 JSON 输出。
+(如果有中英文简历同时出现时，只关注中文简历)
+严格按照 JSON 格式返回结果，不要有任何其他文字。
--- a/rag/prompts/resume_system_en.md
+++ b/rag/prompts/resume_system_en.md
@@ -0,0 +1,3 @@
+You are a professional resume analysis assistant. Your task is to convert the given resume text into JSON output.
+(If both Chinese and English resumes appear, focus only on the English resume)
+Strictly return results in JSON format without any other text.
--- a/rag/prompts/resume_work_exp.md
+++ b/rag/prompts/resume_work_exp.md
@@ -0,0 +1,39 @@
+请从以下带行号索引的简历文本中提取工作经历。
+
+{indexed_text}
+
+提取为 JSON，每段工作经历包含:
+{{
+  "workExperience": [
+    {{
+      "company": "",
+      "position": "",
+      "internship": 0,
+      "start_date": "",
+      "end_date": "",
+      "desc_lines": [start_index, end_index]
+    }}
+  ]
+}}
+
+字段说明:
+- company: 公司全称（含括号内地区信息），如"阿里巴巴(中国)有限公司"
+- position: 职位名称，遵循原文不要编造或推测
+- internship: 该段经历是否是实习，是实习为1，不是为0
+- start_date: 入职时间，格式为 %Y.%m 或 %Y，如 "2024.1"
+- end_date: 离职时间，若至今填写"至今"，若不存在填写""
+- desc_lines: [起始行号, 结束行号]，工作描述对应的行号范围（整数数组）
+  - 指工作经历描述的原文引用段落 index 范围，包括工作成果、业绩、主要工作、技术栈等
+  - 不包括 company、position、start_date、end_date 所在行
+  - 尽可能写全，直到下一段工作经历或其他段落标题为止
+  - 遇到以下段落标题时必须截止，不要将其包含在 desc_lines 中：
+    个人评价、自我评价、个人总结、个人优势、自我描述、技能特长、专业技能、教育背景、教育经历、项目经验、项目经历、证书资质、语言能力、兴趣爱好、求职意向
+  - 如果不存在就写 []
+
+示例:
+[22]: 阿里巴巴 2021.11-2022.11 高级工程师
+[23]: 工作描述: 从事地推工作完成xx业绩
+[24]: 在地推任务中考核为A
+则 desc_lines 应为 [23, 24]
+
+只返回 JSON。 /no_think
--- a/rag/prompts/resume_work_exp_en.md
+++ b/rag/prompts/resume_work_exp_en.md
@@ -0,0 +1,38 @@
+Please extract work experience from the following line-indexed resume text.
+
+{indexed_text}
+
+Extract into JSON, each work experience entry contains:
+{{
+  "workExperience": [
+    {{
+      "company": "",
+      "position": "",
+      "internship": 0,
+      "start_date": "",
+      "end_date": "",
+      "desc_lines": [start_index, end_index]
+    }}
+  ]
+}}
+
+Field descriptions:
+- company: Full company name (including region info in brackets), e.g. "Google Inc."
+- position: Job title, follow original text, do not fabricate or guess
+- internship: Whether this is an internship, 1 for yes, 0 for no
+- start_date: Start date, format %Y.%m or %Y, e.g. "2024.1"
+- end_date: End date, use "Present" if still employed, "" if not available
+- desc_lines: [start_line, end_line], line number range for job description (integer array)
+  - Refers to the original text reference range for job description, including achievements, responsibilities, tech stack, etc.
+  - Include as much as possible until the next work experience entry or other section heading
+  - STOP before these section headings (do not include them in desc_lines):
+    Self-evaluation, Personal Summary, Skills, Technical Skills, Education, Project Experience, Certificates, Languages, Hobbies, Career Objective
+  - Use [] if not available
+
+Example:
+[22]: Google Inc. 2021.11-2022.11 Senior Engineer
+[23]: Job description: Responsible for backend development
+[24]: Achieved 99.9% uptime for core services
+Then desc_lines should be [23, 24]
+
+Return JSON only. /no_think