# How the Mouve food scanner decides what is on your plate

Canonical page: https://mouve.it/blog/how-the-mouve-food-scanner-decides-what-is-on-your-plate
By Mouve team, 9 Nov 2026. 4 min read. Tags: engineering.

Mouve, the fitness and nutrition app made by Mouve Technology OÜ in Tallinn, doesn't ask an AI model how many calories are on your plate. A model looks at the photo and names what it sees. Our own code then finds matching entries in a food database, picks one, and the calories and macros come from that entry.

That split is the biggest design decision in the scanner, so here's how it works.

## What does the scanner do with a photo?

Two steps. First it looks: it checks for a barcode (no AI involved), asks a vision model to name the items and brands it can see, and for packaged food reads the label text. Then it searches: it takes those readings and looks for matches in Open Food Facts, in the products Mouve holds from shops, in a list of common foods and in restaurant menus.

The model never writes a nutrition value into your diary. Our Privacy Policy says the same: the photo is sent to identify the food and estimate portions, and "nutritional values are then calculated from our own food database".

## Why match against a list of known foods?

A model that answers freely can say "chocolate milk" one time and "cocoa drink" the next. A database entry is the same entry every time. Matching against known entries removes that drift, and it lets us show where a number came from.

It helps with size too. A 330 ml can and a 500 ml bottle of the same drink have different totals. A model that has only seen the front of the pack may not know which one it's looking at. A database entry does.

## How does it choose between several matches?

Everything it reads counts as a vote, and some votes count for more. A barcode read straight from the photo is the strongest, and so is a choice you make yourself. Digits printed under a barcode come next, then the text on the label, and last the AI's own guess at names. Extra checks, like pack size, the maker's name or the country the product is sold in, add weight when they agree. A check that disagrees takes nothing away; the product just gets no boost.

A barcode that decodes cleanly, or printed digits that point to exactly one product, settle it. Your own pick always wins.

## A real scan, step by step

From one of our test sessions on 2 October 2026: a 500 g pack of Wõro Õllemops Juustuga with half the barcode hidden.

- 0.6 s: the barcode reader found a barcode but couldn't read it.
- 1.4 s: the AI named the brand and read "ÕLLEMOPS JUUSTUGA" and "500 g".
- 3.4 s: it read the digits under the barcode.
- 5.5 s: the digits matched exactly one product, which settled it.
- 7.4 s: done, with high confidence. The whole scan cost about half a cent.

The right product came from digits printed on the pack, not from a guess at a name.

## Why does it sometimes show choices or ask me?

Because sometimes the evidence doesn't separate. If two products are close, the scanner shows you both instead of picking one and hiding the doubt. If it can't tell which size you have, it lists the sizes it knows and you pick one. If the product isn't in our database yet, it says so and lets you log something close.

Every AI result in the app carries a notice: "AI results are estimates and can be inaccurate. Check the numbers before logging." We mean it. Plates of mixed food are harder than packaged products, because there's no barcode and the portion is a guess.

## What this design can't do

- Name a product that isn't in our database. We keep adding to it; see [how we merge barcode data from many sources](/blog/how-mouve-merges-barcode-data-from-many-sources).
- Weigh your food. Portions from a photo are estimates.
- Tell you what to eat. It reports what an entry says.

We test every change against recorded scans before it ships: [Testing the food scanner with recorded scans, for free](/blog/testing-the-food-scanner-with-recorded-scans). What happens to your photos afterwards is in [where your food photos go and how long we keep them](/blog/where-your-food-photos-go-and-how-long-we-keep-them).