work-cvs

 

One knowledge base, one LaTeX template, and a reproducible build. The point is to stop rewriting a CV from scratch for every posting: PROFILE.md holds everything true about you, a job posting says which parts matter, and cv.tex is the format that turns the selected subset into a one-page PDF an applicant tracking system can actually read.

PROFILE.md  ──┐
              ├──►  tailored cv.tex  ──►  latexmk  ──►  cv.pdf
job posting ──┘

The files

File	What it is
PROFILE.md	The source of truth. Deliberately holds more detail than any one CV should show
cv.tex	The template and the current baseline — style, spacing, section layout
cv.pdf	Committed on purpose: it is the artifact people are actually sent
flake.nix	The LaTeX toolchain, so the build needs no system-wide TeX install

Building

nix develop                          # then: latexmk -pdf cv.tex
nix develop -c latexmk -pdf cv.tex   # one-shot
nix develop -c latexmk -c            # drop the aux files, keep the PDF

A named subset of TeX Live rather than scheme-full — ten packages on top of the small scheme, which is a fraction of the closure and still builds this document unchanged. A \usepackage added to cv.tex may need a line in flake.nix, and the error when it does says exactly which file is missing.

Two of those packages are not named after the file that needs them, which is the only part that ever wastes anyone's time:

• fullpage.sty comes from preprint
• tabularx.sty comes from tools

Note

Nix reads the flake through git, so a newly created file is invisible until it is at least git add -N-ed. error: Path 'x' ... is not tracked by Git means that, not a broken flake.

Why this format, and what "ATS friendly" actually means here

Most CV templates lose text on the way into an applicant tracking system, because the PDF stores glyph indices with no mapping back to characters — the document looks right and extracts as mojibake, or as nothing. Two lines in cv.tex are what prevent that:

\input{glyphtounicode}   % the glyph → Unicode mapping table
\pdfgentounicode=1       % emit it into the PDF

That embeds a ToUnicode CMap, so the text in the PDF is recoverable as text. It is a real mechanism rather than a claim, and it is checkable in one command:

pdftotext cv.pdf - | head -20      # should read as your CV, not as symbols

One place it is imperfect, visible in that same extraction: the contact line renders as GitHub | LinkedIn, so a parser reading text gets the words and not the URLs behind them. Fine for a human, lossy for a machine. If a posting's system is known to harvest links, spell the URLs out instead of hiding them behind \href link text.

The rest of the ATS-friendliness is what the template does not do — no multi-column layout, no text inside images, no icon fonts standing in for words, no header/footer carrying content. Section headings are plain words (Experience, Education, Technical Skills) because that is what parsers look for.

Based on Jake Gutierrez's resume template, itself based on sb2nov/resume. MIT license.

Writing a tailored version

PROFILE.md ends with a CV Generation Notes section — the rules live there rather than here, so that an AI handed only that file still has them. In short: pick the relevant subset, reword freely, invent nothing, keep it to one page, and cut bullets rather than shrinking the margins.

The template's commands, so a tailored version matches the baseline instead of inventing a layout:

Command	Use
\resumeSubheading{a}{b}{c}{d}	A job: company, location, role, dates
\resumeProjectHeading{a}{b}	A project: name + tech, dates
\resumeItem{...}	One bullet
\resumeSubHeadingListStart/End	Wraps subheadings
\resumeItemListStart/End	Wraps bullets under one subheading

Current sections: Summary, Experience, Contributions & Projects, Education, Technical Skills, Languages.

Keep cv.tex as the general-purpose baseline. Cut a version down for one application only if worth revisiting.

Linting

This repo uses markdownlint to keep the Markdown files (PROFILE.md, README.md) consistent. The config is in .markdownlint.json.

nix develop -c markdownlint .

Git hooks

This repo uses git-hooks.nix to run hooks automatically before each commit. The hook configuration lies in flake.nix:

• markdownlint — checks .md files against .markdownlint.json
• latex-compile — runs latexmk -pdf cv.tex on .tex changes to catch compilation errors before they reach the commit

Enter a development shell with hooks enabled:

nix develop

Run all hooks sandboxed (no internet, read-only filesystem):

nix flake check

Run hooks through the dev shell (recommended for formatting hooks):

nix develop -c pre-commit run --all-files

If you're using Neovim, the markdownlint CLI and texlab LSP are both provided by flake.nix via the devShell packages.