Four days ago I started a personal site. It's at 89 pull requests now, and most of them are things nobody asked for.
Half the reason for it is that my profile now often gets read by an LLM before a person. So there are two versions: one for people, with a 3D map of how I build agent systems that you can break, and /llms.txt for models. Both, the PDF CV and the Russian version come from one file.
The first version took two evenings. The rest went into the site paying attention: it works offline, remembers what you read (in your browser only), tells you what changed since your last visit and shows in the footer how long the page took to load for you. Every post grows its own pixel creature, the shadows fall away from your cursor, and Ctrl+P on a post prints a newspaper page.
None of it is hard on its own. What I was really testing is the process: every PR starts with a red test, a writer agent does the work, separate judge agents try to break it, and a performance budget is checked before merge. With all of the above, the site's own JavaScript a page loads up front is still under 8 KB.
Every merged PR, with the test that was red first, what got caught and what I decided, is here: https://rossev.dev/how-its-built/?utm_source=linkedin&utm_medium=social&utm_campaign=site-launch&utm_content=body
Two things that broke on day one get their own posts over the next two days: a logo that regrew its G as an R, and an animation that ran twice as fast on a 120 Hz screen.