GRWM by Kate Noir started as a personal test: could a designer use automation to build a complete product alone? I wanted a portfolio piece that showed product design and AI-assisted coding together, a small launched product rather than static case study visuals.

It is an outfit advice quiz. You pick the weather, the occasion and the company, and get a styling recommendation. For the first release the goal was to prove the concept, with one fixed look for each combination.

I designed the desktop and mobile quiz and the logic behind it: three questions with four answers each, resulting in 64 possible looks. The first result screen layered five transparent garment images per look, which I replaced with one pre-made flat-lay image per combination and a lookup table.

That choice was simpler to build and maintain alone. A universal set of hotspot coordinates lets every look show item tooltips. I also rejected the first AI-generated visual design and created my own, because it was not good enough.

I built the app on Vite and React with a Supabase backend, starting from a Figma Make scaffold and continuing with Claude Code for edits and logic, while still editing code by hand where it mattered. Denys Bendovskiy, an AI product manager, helped with the pipelines and prompts for image generation.

The look images were generated from photo references with an image model, then sorted and named by Claude so they fit the table. Claude also analyzed each image to place tooltips and write the styling tips. The pipeline processed assets concurrently instead of one by one.

GRWM launched in 2026 and is linked from my Instagram. It proved that a designer can build a working, scalable app solo, and the first testers responded positively.

Testers asked for more variety, since each combination currently returns one fixed look. That feedback points to the next iteration: generating a unique image for every result.