Ergograph

Ergograph (Greek ἔργον "work, deed" + γράφειν "to write": "the one that writes down your work") is a YAML-driven CV and dossier generator. From plain content files it produces ready-to-send PDFs: a CV, a project history, a skills matrix and a complete dossier, in any number of languages and variants (e.g. with/without an hourly rate).

The generator contains no personal data. All content and all build steering come from the outside via YAML files; the code only provides rendering, the theme and the PDF export.

How it works

config.yaml + content/<lang>.yaml  ->  HTML (theme "modern")  ->  PDF (Chrome headless)

1. config.yaml steers the build: person, languages, variants, documents, output paths.
2. One content file per language (content/de.yaml, content/en.yaml, …) with all texts, including section labels and document file names.
3. Chrome (headless) renders the HTML intermediate step to A4 PDFs, which then get page numbers and title/author metadata stamped in.

Installation

Requirements: Python ≥ 3.10 and Google Chrome or Chromium. Chrome is only needed for the PDF step (ergograph build --html-only works without it) and is not installed by pip — Ergograph looks for an existing installation (see chrome: below).

# as an isolated tool (recommended)
uv tool install ergograph

# or into the current environment
pip install ergograph

There are no extras to pick: page numbers and the ATS check are always included. Both dependencies (PyYAML and pypdf) are pure Python and together under 1 MB.

Quick start

cd examples/minimal/
ergograph validate          # check config + content files
ergograph build             # build everything (HTML + PDF)
ergograph build --html-only # HTML only, no Chrome
ergograph build --variant mit-stundensatz --lang de

The PDFs end up under pdf/<variant>/<language>/YYYY-MM-DD_<Name>_<document>_<language>.pdf. Older builds are kept side by side thanks to the date prefix (disable it with output.date_prefix: false).

Example output

The rendered example PDFs are committed per persona under examples/<name>/pdf/, e.g. the German CV or the comprehensive architect dossier.

ATS readability

The documents are built to be fully readable by applicant tracking systems: a real text layer (no text in images), reading order equal to content order, skill levels as numbers next to the bars, and PDF title/author metadata. Ergograph verifies this instead of assuming it — after every build it extracts the PDF text layer (pypdf, the same way ATS parsers read PDFs) and asserts that every content string from your YAML appears in it. Findings are reported as warnings; ergograph build --strict turns them into a build failure. Details in docs/SPEC.md (R15/D13).

The steering file config.yaml

person:
  name: Alexandra Argyriou        # appears in the header and in the PDF file names
  # file_slug: Alexandra-Argyriou # optional, default: name with hyphens

theme: modern                     # bundled theme, or path to your own .css
level_max: 6                      # maximum of the skill-bar scale

languages: [de, en]
variants: [mit-stundensatz, ohne-stundensatz]

documents:                        # list (for all languages) or mapping per language
  de: [cv, projects, full]
  en: [full]

content:
  de: content/de.yaml
  en: content/en.yaml

output:
  html_dir: html
  pdf_dir: pdf
  date_prefix: true    # date-stamped file names; false = stable names

# chrome: /path/to/chrome         # optional; otherwise auto-detected

Every field with its type, default and whether it is required: docs/CONFIGURATION.md.

The content files

One YAML file per language holds everything that appears in the documents. The picture shows which key renders which region — red comes from content/<lang>.yaml, green are the headings from labels, and blue is the one visible field that comes from config.yaml:

Three keys are not visible above: projects builds the project history, skills the skills matrix, and doc_names supplies the PDF file names. All fourteen keys are required, but every list may be empty — publications: [] simply drops that section, which is how one format serves a software architect and a carpenter.

Variants are steered declaratively: a fact carrying variants appears only in those variants, everything else appears everywhere.

facts:
  - label: Availability
    value: immediately
  - label: Hourly rate
    value: €100/h
    variants: [mit-stundensatz]

Content values are trusted HTML fragments: write UTF-8 directly (ü, €, "…"), and use <a href="...">…</a> for inline links where needed.

Full field reference — every field, its type, whether it is required and its default: docs/CONFIGURATION.md. The repository also ships JSON Schemas for both file types, so an editor can complete fields and flag typos while you write.

Generating content with an AI agent

The content files are plain, flat YAML with every text outside the code, which makes them a practical target for an agentic coding tool (Claude Code, Codex, Cursor, …): the agent gets a machine-readable contract in schemas/, working examples, and two commands that judge the result objectively — ergograph validate for the structure and ergograph build --strict, which fails when a string from the YAML is not extractable from the finished PDF. That turns "write my CV data" into a loop the agent can close by itself.

A ready-to-use prompt, plus what to check by hand afterwards (numbers, dates, certificate titles — a model formats reliably and invents plausibly), is in docs/CONFIGURATION.md.

Examples

Every example under examples/ is a fictional persona and ships with its rendered PDFs:

Variants are steered declaratively: an entry in facts with variants: [mit-stundensatz] only appears in that variant, all other facts appear everywhere.

facts:
  - label: Availability
    value: immediately

Content values are trusted HTML fragments: write UTF-8 directly (ü, €, "…"), and use <a href="...">…</a> for inline links where needed.

Development

uv run pytest                    # test suite, no Chrome and no network needed
uvx pip-audit -r <(uv export --format requirements-txt --all-extras --no-dev --no-emit-project)
trivy fs --scanners vuln,secret,misconfig .

Every push runs the suite on Python 3.10–3.14, renders all examples, and scans dependencies and sources (pip-audit, Trivy, CodeQL). Releases go to PyPI from a v* tag via trusted publishing. Version history: CHANGELOG.md.

License

MIT