Adra · Personal recognition
It learns from you.
Your home cooking is not in any restaurant database. So Adra learns it, from you, on your phone. It starts general, and it becomes personal one saved dish at a time.
Save once, matched later
Recognition begins with the catalog, and the catalog is general. It knows a plate of pasta, but not your version of it, and not the way you portion it. When you save a dish, Adra keeps a small image fingerprint of its photo on your device, right next to the dish itself.
The next time you photograph something similar, your saved dish can match first, before anything generic. The app does not start out knowing your kitchen. You teach it by saving.
Save a dish once, and Adra knows it from then on.
Under the hood
1. You save a dish → its photo becomes a 512-number image fingerprint 2. You photograph it again → the fingerprint is compared with your saved dishes first 3. A strong match → your saved values return before any catalog search
The comparison is image against image, on the phone, which is why a match does not depend on what the dish is called. No name matching, no cloud, no round trip. When nothing matches strongly enough, the normal recognition path runs unchanged, so a new dish is treated like any other food.
The fingerprint is what makes the match possible. Your saved values are what make it useful.
Your corrections teach it
A correction is not a chore. It is the moment the app learns what you meant, and the moment it gets sharper for the next photo. Confirm a match and the suggestion holds. Fix one and the fix is remembered.
All of this happens on your phone, for you. It never trains anyone else's system, and nothing is uploaded.
Everything stays yours
The fingerprints, the corrections, your own dishes: all of it lives in your local database, next to the rest of your diary. There is no account, so there is nothing to sync and nothing to hand over.
Your food knowledge is yours in the plainest sense, because it never leaves the phone that made it. The privacy page has the full story.
In time, your dishes count as foods
You can add your own dishes and recipes, and in time they are detected like the rest of the catalog. That phrasing comes from the app's Play Store listing, and it is deliberately careful, because this part is early.
Personal recognition grows with every dish you save. It is honest work in progress, not a finished feature.
Honest limits
-
Matching is fuzzy at first.
Lookalike dishes can confuse anyone, including Adra. A wrong match is not a dead end. The Fix it button corrects it in one tap, and the correction is remembered.
-
Dishes without a photo stay name-searchable.
A favorite saved without a photo cannot be matched from the camera. It stays searchable by name, and adding the photo later opens the camera path.
-
This is early work.
Personal recognition grows with use. It is honest work in progress, not magic, and it is already useful for the dishes you cook often.
FAQ
Is my data used to train a model?
No. Matching compares your photos to your own saved dishes, on your device. Nothing is uploaded and nothing is shared.
How does Adra recognize my own dishes?
It compares the new photo's image fingerprint with the fingerprints of the dishes you saved. A close match comes back with your saved values.
What if the match is wrong?
Use the Fix it button. The correction is remembered, and the suggestion gets sharper next time.