AI / Mobile·2026
Synvora: AI skincare app
Consumer skincare app with camera-based vision analysis and an LLM recommendation engine. Built solo as technical cofounder.
Empty repo to App Store launch in under four months
- React Native
- Expo
- TypeScript
- Supabase
- Claude API
- RevenueCat
The product
Synvora is a consumer skincare app for iOS and web. You scan your face from three angles, and it gives you observations and a routine built from products you can actually buy in Canada. It's live on the App Store.
What I built
I built the whole product as the technical cofounder:
- A React Native (Expo, TypeScript) iOS app with 39+ screens and subscription billing through RevenueCat.
- A Next.js marketing site.
- A multi-stage LLM backend on Supabase Edge Functions: a three-angle vision scan, a cached verdict layer, a routine assembler, a safety gate, and a conversational coach that remembers prior scans and onboarding answers.
Decisions that mattered
- Routing by task to cut cost. Claude Sonnet handles vision analysis and product verdicts; Haiku handles chat. Verdicts are cached against a curated product database instead of re-querying per user.
- Observations, not diagnosis. The model is constrained to describe visible traits only. That product-safety rule shaped the prompt contract and the App Store health-content rating.
- A curated product database. 108 Canada-available products with metadata like fragrance-free flags, ingredient conflicts and pregnancy cautions, merged and cleaned from Open Beauty Facts and enriched with the Claude API.