Grimoire Culinaire
Private BetaFounder, Product Owner, Architect, AI Solutions
AI-enabled Android mobile recipe management app, built as a real commercial product. Flutter / Dart 3-layer Riverpod architecture, offline-first persistence via Drift / SQLite, cloud sync through Supabase / PostgreSQL + Firebase Analytics. Multi-modal AI capture (voice, text, photo, URL) with multi-provider LLM routing (OpenAI API primary + Gemini fallback). Internal economy (muffins) with immutable ledger, 3-tier freemium via RevenueCat, free-tier monetization via AdMob. i18n FR / EN / ES / IT.
What this project proves
Private Beta Android · 14 lots · 400+ tests
Current state
Private Beta • 2025
Design focus
Context
Personal project built as a complete commercial product. The goal: prove a solo developer can ship an AI-enabled app, from architecture to Google Play private beta, using a multi-LLM routing as a product accelerator without ever losing product ownership.
Challenges & constraints
Making AI actions useful in daily use without breaking reliability or the offline-first experience. Building an AI governance surface (per-user quotas, plans, promos, admin ledger, pilots) operable by non-technical admins. Designing an internal economy balanced between freemium conversion and retention. Holding 80%+ test coverage across 14 successive delivery lots.
Results & learnings
Complete MVP in private beta on Google Play. 400+ tests, 80% coverage. 14 lots shipped without major tech debt, 7 Supabase Edge Functions, 51 analytics events, 18 AI actions with quotas and fallback. Riverpod 3-layer + offline-first pattern reusable on other projects. AI governance surface operable by non-technical admins.
Tech Stack
Features
- ▸Multi-modal AI capture (voice, text, photo, URL)
- ▸Multi-LLM routing (OpenAI API + Gemini fallback)
- ▸3-tier freemium economy (Free / Plus / Pro)
- ▸Gamification (XP, streaks, achievements)
- ▸Cooking mode with timers
- ▸PDF export + social sharing
- ▸Offline-first Supabase sync
- ▸i18n FR / EN / ES / IT
- ▸AI governance surface (quotas, plans, ledger)
Related paths
Links
Production note
The most structural decision was treating the internal economy (muffins) as a real monetary system from lot 1, not as a late addition. It conditioned the entire architecture — reserve before debit, rollback on failure, immutable ledger. Same for the AI governance surface: designed from the start as a product for non-technical admins, not as a dev dashboard. Without those two choices, the 14 lots would have been much more painful.