Live
2019 – present
EmptyMyFridge
AI-driven anti-food-waste app
Type in the ingredients you already have at home and get recipes that use them. I founded it, built it from scratch and still maintain it: 40k+ downloads, about 150 daily users, 4.5★ on iOS and 4★ on Android.
- Full product ownership, from idea and design to store releases and support
- React Native New Architecture, Firebase Cloud Functions back end
Stack: React Native, TypeScript, Firebase, Node.js, iOS, Android, Web
WebsiteApp StoreGoogle Play
Live
2026 – present
Sport Challenge
Thirty body-weight exercises, played as a ladder of levels
A home-workout app built like a game: pass the current level to unlock the next, then read the sourced health benefit of what you just did. Skip three days and the last level locks again. Everything stays on the phone: no account, no server. Live on the App Store and Google Play.
- Native look on both platforms through @expo/ui (SwiftUI on iOS, Jetpack Compose on Android), Reanimated 4 animations, expo-sqlite storage
- Tests, type-check and lint on every pull request; EAS release workflow with over-the-air updates on every push to main
Stack: Expo SDK 57, React Native, TypeScript, Expo Router, @expo/ui, Reanimated 4, EAS
App StoreGoogle Play
Work in progress
2026 – present
Korean Challenge
Learn Korean as a map of worlds to unlock
The Sport Challenge concept applied to language learning: a ladder of Korean levels drawn as a map of worlds, each one a timed round of nine kinds of question (recognise, produce, listen, build, sprint, no mistakes) drawn from short lessons. Three days away and the last level locks again. French or English interface, Korean either way, all on-device.
- Curriculum, pedagogy and ladder rules as pure TypeScript modules with tests; generated speech assets for the coach
- Same native stack and release pipeline as Sport Challenge, plus headless-browser probes to review many screens at once
Stack: Expo SDK 57, React Native, TypeScript, Expo Router, @expo/ui, Reanimated 4, expo-speech, EAS
Work in progress
2026 – present
Locale
A private AI that runs on your computer and says when anything goes online
A desktop app that looks at your machine (memory, GPU, disk), recommends the best open model for what you want to do, downloads it once from Hugging Face and then runs it with no internet. A light at the top of the window shows at all times whether anything is leaving the computer; airplane mode blocks every outgoing connection and a network log lists each one.
- Network guard wrapping fetch, http and https in the main process; the window itself cannot load anything online
- Model recommendation from available GPU memory and a scored catalogue of 14 open models, checked against Hugging Face
Stack: Electron, node-llama-cpp, llama.cpp, Metal, JavaScript, Hugging Face
Work in progress
2025 – present
Histoire du monde
A 715-entry timeline from the Big Bang to today
A long-form personal research project: a chronology in 14 chapters, from cosmology to the present day, each entry sourced and illustrated. I built the tooling around it too: image conversion to WebP, media hosting on Cloudflare R2 and video generation. Not published yet; it is being reworked before going back online.
- 715 dated entries, 14 chapters, hundreds of captioned illustrations
- Custom scripts for media optimisation and upload
Stack: Writing, Data curation, Static site, WebP pipeline, Cloudflare R2