resume-parse

name	resume-parse
description	Extract structured data from PDF or DOCX resume files. Produces a ResumeData JSON object with sections for personal info, education, experience, publications, skills, projects, and certifications. Use when processing an uploaded resume file to convert raw document content into structured data.

Resume Parse

You are tasked with extracting structured resume data from raw text extracted from a PDF or DOCX file.

Input

You receive the raw text output from the extract-text script. This text is a best-effort extraction and may have formatting artifacts.

Output

Return a JSON object matching the ResumeData schema:

{
  "name": "string",
  "email": "string | undefined",
  "phone": "string | undefined",
  "location": "string | undefined",
  "links": [{ "label": "string", "url": "string" }],
  "summary": "string | undefined",
  "experience": [{
    "title": "string",
    "company": "string",
    "location": "string | undefined",
    "startDate": "string",
    "endDate": "string | undefined",
    "bullets": ["string"]
  }],
  "education": [{
    "degree": "string",
    "institution": "string",
    "year": "string",
    "gpa": "string | undefined",
    "details": ["string"]
  }],
  "skills": ["string"],
  "certifications": ["string"],
  "projects": [{
    "name": "string",
    "description": "string",
    "technologies": ["string"],
    "url": "string | undefined"
  }],
  "rawSections": [{
    "title": "string",
    "content": "string",
    "startLine": 0,
    "endLine": 0
  }]
}

Extraction Rules

1. Personal Info: Name is usually the largest/first text. Email, phone, location, and links follow. Look for patterns like mailto:, +1-, URLs.

2. Education: Look for keywords like "Education", "Academic". Each entry has institution, degree, date range, and optional GPA/details.

3. Experience: Look for "Experience", "Work", "Employment". Each entry has company, title, date range, location, and bullet points.

4. Publications: Look for "Publications", "Papers", "Research". Each entry is a single citation string.

5. Skills: Look for "Skills", "Technologies", "Technical". May be comma-separated or categorized (e.g., "Languages: Python, Go").

6. Projects: Look for "Projects". Each has name, description, and optional technologies.

7. Certifications: Look for "Certifications", "Licenses".

rawSections — Complete Section Inventory

The rawSections array is CRITICAL for template filling. It must capture every section in the resume, including non-standard ones:

• Always include ALL sections found in the resume, even if they don't map to a structured field (experience, education, skills, etc.).
• Common extra sections to capture: Summary, Objective, Awards, Honors, Certifications, Volunteer Work, Leadership, Languages, Interests, Activities, References, Professional Affiliations, etc.
• Each rawSection entry preserves the original section title and full content.
• The content field should contain the complete text of the section as it appears in the resume — do not summarize or truncate.
• rawSections serve as the source of truth for sections that don't map to structured fields, enabling the template filler to create new sections.

Edge Cases

• Empty sections: If a section has no content, omit it (empty array).
• Non-standard headings: Use context clues to identify sections.
• Multi-language: Handle both English and Chinese resumes. CJK names may appear alongside English names (e.g., "张伟 (Wei Zhang)").
• Merged sections: Some resumes combine Work and Projects. Split them into the appropriate sections based on content.
• Date formats: Normalize to human-readable format. Accept "2020-01", "Jan 2020", "2020.01", "January 2020" etc.
• Non-standard sections: Always capture in rawSections even if they don't fit any structured field. Examples: "Volunteer Work", "Awards", "Professional Development", "Leadership", etc.

For detailed format-specific edge cases, consult references/FORMATS.md.