7 Sept 2026 · meta, agents
How this site works
This site is deliberately boring underneath and slightly silly on top.
The boring part
It’s Astro, building to plain static HTML, hosted on GitHub Pages. There is no backend, no database and no CMS. Everything is a file in a git repo:
- My CV lives in one YAML file, validated at build time. If I typo a key, the build fails instead of quietly dropping a job off my resume.
- Notes are Markdown files in a folder. Adding a note means adding a file.
The part that’s actually the point
Most personal sites are written for one audience — people with a browser. This one is written for two, because increasingly the thing reading about you is a language model deciding whether to mention you.
So every page here exists in three forms:
- HTML, for people.
- Plain Markdown, at the same URL with
.mdappended —/cvbecomes/cv.md. No navigation, no styling, no token tax. - Structured data, as a JSON-LD
Persongraph in every page’s<head>, so a machine can extract “where does she work, what does she know, what else is she” without parsing prose.
There’s also an /llms.txt index listing every page with a
one-line description, which is a convention for exactly this. I stole the
pattern from Taivo, who is my manager and who
did it first.
The Stargate
The notes on this site sit behind a seven-chevron dialling puzzle. The hints are
on the page. It is not security and it is not access control — the notes are
right there in the HTML, in the sitemap, in llms.txt, and at their own URLs.
Anything fetching this site without running JavaScript sees everything
immediately and never knows there was a gate.
The gate is a toy for humans who want one, and there is a link to skip it for humans who don’t. Making a puzzle that blocked crawlers would have been easy and would have defeated the entire point of the site.
