Aaryn Biro Portfolio

Professional static HTML portfolio for Aaryn Biro.

The live page source is page/page.html. scripts/build_site.sh injects SEO metadata and gathers the supporting assets into a deploy directory - site/ by default, or docs/, which is the copy GitHub Pages serves (see Deployment).

Resume PDF

The downloadable resume is built from a Word document, not from RESUME.md:

File	Role
assets/Aaryn_Biro_Resume.docx	source of truth - design and text
assets/github_stats.json	quarterly feed, refreshed by scripts/fetch_github_stats.py
assets/Aaryn_Biro_Resume.pdf	what the site serves
assets/resume.lock.json	hashes of the three above, written by the build

bash scripts/build_resume_pdf.sh

That resolves the {{GITHUB_CONTRIBUTIONS}} placeholder in the DOCX from the stats JSON, converts with LibreOffice, and rewrites the lock.

It runs locally, not in CI. The PDF's line breaks depend on the fonts of the machine doing the conversion, so rebuilding it on a Linux runner would reflow the page. scripts/check_resume_lock.py runs in CI instead and fails when the DOCX or the stats JSON changed without a rebuild - including after the quarterly stats workflow commits, which is the signal to rebuild locally and commit the PDF.

RESUME.md is a separate, shorter document for the web page and for agents. It is not the source of the PDF: it has a Selected projects section the PDF does not, and the PDF has Education, Certifications & Speaking and a phone number that RESUME.md does not. An earlier build_resume_pdf.sh ran pandoc RESUME.md -o assets/Aaryn_Biro_Resume.pdf, which would have replaced the designed one-page resume with a plain pandoc render the moment pandoc appeared on a runner.

Agent-ready metadata (L6)

Dual-audience portfolio for humans and LLM agents (see pxl-registry S5 L6):

Asset	Path	Live URL
llms.txt	repo root → site/llms.txt	https://aabiro.github.io/llms.txt
Agent manifest	.well-known/agent.json	https://aabiro.github.io/.well-known/agent.json
JSON-LD	scripts/inject_seo.py	Person + CreativeWork @graph
No-JavaScript copy	scripts/inject_static_profile.py	<main id="static-profile"> in index.html
Resume source	RESUME.md	built to assets/Aaryn_Biro_Resume.pdf by scripts/build_resume_pdf.sh

The Claude Design export renders only with JavaScript, so on its own a crawler, link previewer, recruiter tool or agent fetching the page sees an empty loader. scripts/inject_static_profile.py reads the page's own copy and component data (experience, work, capabilities, stack, contact) out of the bundled template and writes it into index.html as plain HTML. Browsers with JavaScript still swap in the designed page as before. Re-exporting page/page.html keeps the static copy in sync automatically; the build fails if it can no longer find the experience or work data.

scripts/build_site.sh copies agent files into the deploy directory alongside the HTML page.

Quarterly GitHub contribution stats: .github/workflows/resume-quarterly.yml.

Deployment

GitHub Pages serves the site from the committed docs/ directory on main, so deploying is git push. Rebuild docs/ whenever page/, web/, llms.txt, .well-known/ or the resume PDF changes, and commit it alongside the change:

SITE_OUTPUT_DIR="$PWD/docs" bash scripts/build_site.sh

This used to deploy through Actions, with actions/upload-pages-artifact and actions/deploy-pages. That path is blocked: the upload needs Actions artifact storage, which this account's billing has cut off, so the job failed at the upload step with "Artifact storage quota has been hit" after an otherwise green build. Serving docs/ from the branch needs no artifact storage and no hosted runner, and keeps the same URL.

.github/workflows/github-pages.yml verifies rather than deploys: it rebuilds the site and fails if the committed docs/ differs from a fresh build, which is the only thing that catches a stale docs/.

GitHub-hosted runners are blocked on this account too, so that check runs on the sandboxed self-hosted runner from aabiro/xcelsior (scripts/ci-runner/). Start one from the xcelsior checkout; it serves the one job and exits:

XCELSIOR_CI_REPO=aabiro/aabiro.github.io ./scripts/ci-runner/run-runner.sh

A push with no runner up just queues the check. The site is live either way.

The workflow has no pull_request trigger on purpose: on a public repository, that trigger would let anyone's pull request run code on the self-hosted runner.

Check out the live site at:

https://aabiro.github.io/