Consumer mobile · AI · 2026
Matcha
AI color analysis for fashion — find out which colors actually flatter your skin tone.
Problem
Color analysis has been gatekept behind expensive in-person consultations and vague online quizzes. Most people own clothes that fight their skin tone without knowing it.
Insight
If a phone camera can read skin tone accurately, the rest of the experience can be radically simpler than a quiz — point, capture, and get an honest answer in seconds.
Solution
A mobile app that analyzes a user's skin tone from a single photo and recommends a personal color palette, then lets them check any item in their closet against it.
Architecture
Native iOS (Swift) and Android (Kotlin) clients with Supabase as the backend of record. On-device image preprocessing, then edge functions for color extraction and palette generation, with OpenAI used for nuanced recommendation language. Rate limiting and RLS protect both cost and user data.
Outcome
Shipped V1 to the App Store and Google Play in 2026 with a focused color analysis flow, a closet checker, and a private profile. Early users describe it as 'the first quiz that didn't feel like a quiz.'
What I wanted to build
I wanted Matcha to feel less like a quiz and more like having a calm, well-read friend look at your closet and tell you the truth. Most color tools online are either too academic (twelve seasonal subtypes, charts, vocabulary tests) or too shallow (“you are a winter, here are sparkles”). Neither makes anyone a better dresser.
The bet was that a phone camera, an honest analysis, and a tight feedback loop on real clothes could do more for someone than a hundred quizzes.
The hardest parts
Color is unforgiving. The same skin tone reads differently under warm indoor light, daylight, and the bluish cast of a bathroom mirror. Most of the engineering work was not in the model calls — it was in the preprocessing pipeline that decides whether a photo is even good enough to analyze, and in being willing to ask the user to retake instead of pretending.
The second hard part was language. Telling someone their season in jargon is useless. The copy had to translate “cool, soft, light” into “this olive jacket is fighting you” without sounding like a horoscope.
What’s next
The closet checker is the wedge. Once people see a tool that tells them the truth about clothes they already own, the rest of the product — recommendations, outfit checks, shopping filters — unlocks naturally. V2 focuses on memory, history, and shared palettes.