FoodCheck
A grocery app for the store aisle that uses AI to identify products and rate them for my goals.
Try the real app
This is FoodCheck itself, loaded with sample products. Things to try:
- Search, or filter by Good, Bad, and star rating
- Open a product to see its rating and why it scored that way
- Tap Check to rate something without saving it
FoodCheck is the grocery app I use in the store. It keeps a shared list of products marked good or bad, so the answer is one search away in the aisle.
Snap the front of a pack and AI identifies the product and fills in the nutrition. Scan a barcode and it pulls the data from Open Food Facts. Every product gets a one-to-five-star rating with the reasons spelled out, and a quick check rates something without saving it.
In the aisle, the answer has to come back fast, so Claude Sonnet reads the label in about six seconds. Then a more powerful model, Claude Opus, checks the numbers on its own. Within minutes, it rereads the label, compares the maker’s and retailers’ figures when it needs to, and sends back any fix, which the app applies by itself.
Before switching the in-store reader to Claude, I tested it on 14 real products whose labels had already been verified. It matched 75 of 84 nutrition values, up from 57 with the old setup. Both models run on the API credit that comes with my Claude plan, with a hard monthly cap, so the app can’t run up a bill.
The demo is the real app with sample products. Syncing and AI are switched off in the demo, and your changes stay on your device.
What I built
- Photo recognition with Claude Sonnet (or free Gemini), plus barcode scanning
- An automatic deep check: Claude Opus rechecks every new product against its label within minutes
- A live test that scores the reader against verified labels before changes ship
- Star ratings based on protein and fiber per calorie, with penalties for sugar, saturated fat, and salt
- Works offline and syncs between phones through a private GitHub repository
- One file, no server, and no build step