Mealy plans a household’s week of lunches and dinners around its budget, diet and supermarket, with a shopping list priced against real products. It is built for the Netherlands and Belgium, in Dutch. I owned product direction, UX/UI, architecture and release decisions, and reviewed every change. Implementation was AI-assisted, with Claude Code as co-author on 119 of 120 commits.
Mealy’s catalogue holds 500+ recipes, linked to 5,000+ real supermarket products and prices, so planning a week and pricing its shopping list happen in one flow.
Status: version 1.1.0 is approved by Apple and awaiting manual release.
Tech: Expo SDK 54 (React Native, TypeScript), Zustand, Supabase (Postgres, row-level security, edge functions), RevenueCat and StoreKit, EAS Build, Vercel, and PGlite for database tests.
The challenge
Dutch households plan dinners around a weekly grocery budget and the supermarket they already use, yet recipe apps rarely know what a week will cost there. Mealy answers “what do we eat this week, and what will it cost at our store?” in a single flow.
My approach
Plan on the device, keep content on the server
The planner runs locally, so a week appears instantly. It filters out allergies, diets, disliked foods and missing kitchen equipment, then balances taste and variety across meal concepts so a week does not serve the same dish twice in different forms. Recipes, prices and photos load from Supabase, which means content can improve without an app update or a new review.
Prices people can trust
Each ingredient maps to a real supermarket product, with prices from 13 Dutch chains. Prices older than a week are refreshed in the background, and where no live price is available the app falls back to a marked estimate. Shelf and promotion prices are kept apart, so a budget never rests on a temporary deal.
Honest savings and subscriptions
I replaced an unsourced “you save X” formula with a cited Dutch food-waste statistic, presented as an average. Mealy offers monthly and yearly plans with a seven-day trial, one full plan-shop-cook cycle. Prices come straight from the App Store, the amount actually charged is the most prominent number, and declining the paywall still lets people into the app.
Key challenges
Dinners that did not feel like dinner
TestFlight testers praised the design but flagged odd dinners and unbalanced weeks. I traced the causes to the app rather than the AI model: lunch dishes were leaking into dinner, and some ingredients were missing from the shipped build. I fixed both, then shifted from prompt-tuning to a curated catalogue of meal concepts, with nutrition and dietary labels derived from ingredients.
Polishing under release pressure
In the final days, testers found a clipped savings figure, outdated legal links and an unclear wait after payment. All three were fixed in the approved build, and manual release lets me choose the launch moment.
Outcomes
Verified
- Version 1.1.0 approved in App Review, with subscription purchases tested end to end in Apple’s sandbox.
- 530 passing automated tests and 58 versioned database migrations, all executed against an in-process Postgres.
Qualitative
- TestFlight feedback was positive on design, photography and instructions, and its criticism of recipe choice directly shaped the planner and the catalogue.
