GetJobAI Resume Template

Typst resume template with three visual styles. ATS compatibility first, aesthetics second.

Requires typst 0.13+ (developed on 0.14.2).

This repository is the development home of the template: the test files and Justfile here exist to iterate on template.typ in isolation and to produce the figures used in the project paper. template.typ is the single deliverable — it is copied verbatim into the pdf-generator service repo, which compiles it in-memory from JSON resume data and serves the resulting PDF over HTTP. Nothing else in this directory ships; only template.typ crosses the boundary.

Files

File	Purpose
template.typ	Main deliverable — exports resume(data)
tests/*.typ	Self-contained test files; each imports template.typ and calls resume(data)

template.typ is a library: it exports a single resume(data) function. Test files in tests/ are the entry points for local dev. In production the pdf-generator service builds the data dictionary from resume JSON and calls resume(data) directly.

Style priority: --input style=X > data.style > "professional". The service sets data.style; the CLI --input flag overrides it for local dev.

Styles

Style	Font	Accent	Target
professional (default)	Libertinus Serif	Navy #1a3a5c	Corporate, Consulting
minimal	New Computer Modern	Black	Academic, Research
technical	JetBrains Mono	Steel #005f87	Engineering, Open-Source

Usage

Test files are the entry points. Compile with --root . so Typst can resolve ../template.typ:

# Compile a specific test file
typst compile --root . tests/default.typ pdf/default.pdf

# Override style at the CLI (overrides data.style in the file)
typst compile --root . tests/default.typ pdf/default.pdf --input style=minimal

# Watch
typst watch --root . tests/default.typ pdf/default.pdf

Or use the Justfile:

just compile tests/default.typ # compile one file
just watch tests/carpenter.typ # watch one file (defaults to tests/default.typ)
just all                       # compile all tests/*.typ
just all-styles                # compile every test in all three styles
just check                     # compile all + verify every font is embedded
just clean                     # rm -rf pdf/

What's Been Done

ATS parsing:

• Single-column layout throughout — no multi-column, tables, or floating objects
• h(1fr) for right-aligned dates: looks two-column visually, but is a single text stream in the PDF (verified with pdftotext -layout)
• Contact info in the document body, not in PDF header/footer regions (parsers often skip those)
• Standard section names: Experience, Education, Skills, Certifications, Languages, Projects
• box() around contact URLs prevents mid-path line breaks (github.com/ + newline + janedoe)
• No icon glyphs — plain Unicode separators only
• Fonts are embedded by default (Typst guarantees this); Libertinus Serif and JetBrains Mono both cover the Cyrillic Unicode block for Cyrillic institution names

Data handling:

• All contact fields except name are optional and omit cleanly
• Education grade is a free-form string — supports "5.0 / 5.0" (Ukrainian 5-point), "1.3 (DE)" (German inverted scale), "94 / 100", etc.
• Experience company and title are optional — omit both for career gap entries
• hide: true on any experience or education entry suppresses it without removing it from the data (useful for per-application tailoring in the diff view)
• Bullet points are Typst content blocks [like this], so inline *bold* works naturally for tech keyword highlighting

Localisation:

• Section headings are overridable per-resume via data.headings — pass any subset; unset keys fall back to English defaults
• Sample data uses MM.YYYY date format
• Language level is rendered verbatim as name — level, so the caller controls the wording — CEFR codes (C1), full text (Professional Working (C1)), or Native all work. The pdf-generator service normalises LinkedIn proficiency strings to CEFR before sending

Diff view:

• #diff-added[…] and #diff-deleted[…] are defined as green-underline / red-strikethrough — no-ops unless a diff tool injects them, compatible with typdiff marker style

What's not Included

• Colour toggle — no flag to strip accent colours for a plain-black ATS submission; all three styles render in colour
• Photo field — not planned for MVP; the Export Service has no image pipeline
• Single-page enforcement — no hard clip at page 1; content overflows naturally (correct for German CVs, which expect 2–3 pages)
• Formal ATS testing — pdftotext -layout passes cleanly; ATS tests pass with at least 80/100 with the other 20 points being subtracted for contents.

Known quirk: In the professional style, pdftotext without -layout may split "Jane Doe" across two lines due to the large bold font's word spacing. With -layout (and in real ATS parsers that read the PDF content stream directly) it's a single text run and parses correctly.

Data Schema

#let data = (
  style: "minimal",   // optional — "professional" | "minimal" | "technical"
  headings: (         // optional — override any subset of section headings
    summary: "...",
    experience: "...",
    education: "...",
    skills: "...",
    certifications: "...",
    projects: "...",
    languages: "...",
  ),
  contact: (
    name: "...",      // required
    email: "...",     // optional
    phone: "...",     // optional
    location: "...",  // optional — metro area, not street address
    linkedin: "...",  // optional — without https://
    github: "...",    // optional — without https://
  ),
  summary: [...],     // optional — Typst content block

  experience: (
    (
      company: "...", // optional — omit for gap entries
      title: "...",   // optional
      dates: "MM.YYYY – MM.YYYY",
      location: "...", // optional
      bullets: (
        [Bullet text with optional inline *bold*.],
      ),
      hide: false, // optional — suppresses entry when true
    ),
  ),

  education: (
    (
      institution: "...",
      degree: "...",
      dates: "MM.YYYY – MM.YYYY",
      location: "...", // optional
      grade: "...",    // optional — free-form string
      hide: false,     // optional
    ),
  ),

  skills: (
    (category: "...", items: ("...", "...")),
  ),

  certifications: (
    (name: "...", issuer: "...", date: "MM.YYYY"),
  ),

  languages: (
    // level: rendered verbatim — CEFR code "C1", full text, or "Native"
    (name: "...", level: "..."),
  ),

  projects: (
    (name: "...", description: [...], url: "..."),  // url optional, without https://
  ),
)

ATS Extraction Check

Inspect the extracted text layer to confirm parsers see clean, linear output — section headings in order, dates on the same line as company names, no scrambled text:

pdftotext -layout pdf/default.pdf -

just check is the automated guard: it compiles every test file and fails if pdffonts reports any unembedded font, since a missing font breaks text extraction. The pdf-generator service ships the same recipe.

Credits

Structure and spacing approach inspired by:

• clickworthy-resume by Abdullah Hendy (MIT)
• guided-resume-starter-cgc (MIT)