Yeah it was interesting to me that I identified it immediately.
Perhaps been spending too much time w/ Claude
It's capable of many different designs but this site tells me Rashid didn't even bother.
Use proven traditional statistical methods like supervised learning for this. You can use traditional (not large) language models to tokenize the content and then a supervised learning trained on the most popular models to detect if those models generated the content.
That style is still rare enough that I prefer it over OP's "we installed shadcn + Tailwind and look like every seed-stage startup SaaS from 2024" style.
Or since a lot of vibe coded static sites are just plain html, cs, js, a first try attempt via wget to mirror the first 500KB of the site and then analyze it.