This site is powered by a structured knowledge base (a JSON file that tracks every report, dashboard, code project, and media appearance I've produced since 2019). I built it by transforming my CV into structured data, then enriching it with web searches that found publications and mentions.
That database is what the Research and Lab pages pull from. The semantic analysis in the footer runs on top of it. Add an entry to the source data, run a sync script, and the website updates.
The trade-offs of AI-assisted development
I used AI coding assistants (Claude and Gemini) throughout. The knowledge base took about a week; the website about two weeks. That timeframe isn't impressive on its own, but it's interesting because of the trade-offs involved.
On one hand, I was able to work with technologies I'd never touched before — Astro, Svelte, Tailwind v4. I couldn't pass a test on any of them, but I now have enough understanding to reach for them when I need them. That's a real outcome.
On the other hand, the process got messier as the codebase grew. More files to track, more context for the AI to lose, more cleanup accumulating. Vetting changes takes longer than it should. There's still work to do on refining how I use these tools.
What I'm still figuring out
This is functional but not polished. Some animations break. Some mobile layouts need work. The footer visualization is an experiment that may or may not add value. I'm treating v0 as a starting point, not a finished product.