How the sites are made
Every site on this page came out of design-crawler, a pipeline I wrote in Python. It studies award-winning websites, keeps what they do well in a library, and builds new sites from it. An AI model writes the code inside the pipeline. I write the pipeline, set the brief, read every review and decide what ships.
The three I like most are ARDEI, STILLE and VASCA. What they share: the subject’s own material is the picture, and one drawing engine is reused everywhere on the site.
- 01
Discover
Reads the Awwwards Site of the Day listings into a queue with their scores and tags.
- 02
Capture
Opens each site in headless Chromium: screenshots while scrolling, the phone view, hover and menus, and the site’s own code.
- 03
Analyse
An AI model reads each capture and writes down the type, colour, layout and motion it finds.
- 04
Distil
Merges the analyses into a design library: principles, pitfalls and 61 runnable recipes with measured timings.
- 05
Build
Writes a new site from a brief and the library, with an AI model writing the code.
- 06
Review
Tests the site in a real browser at three sizes, a critic scores it, and I rate it. Then the next round.