Adra · Personal recognition
It learns your food.
Your home cooking is not in any restaurant database. So Adra learns it. From you, on your phone.
Save once, matched later
When you save a dish, Adra keeps a small image fingerprint of its photo, right on your device. The next time you photograph something similar, your saved dish can match first, before anything generic.
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. No name matching, no cloud. If nothing matches strongly, the normal recognition path runs unchanged.
Your corrections teach it
Confirm or correct a match and the suggestion sharpens for next time. The learning 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 in your local database, next to the rest of your diary. No account is involved. 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. This part is early. It gets better with every dish you save.
Honest limits
-
Matching is fuzzy at first.
Lookalike dishes can confuse anyone, including Adra. A wrong match is fixable with the Fix it button.
-
Dishes without a photo stay name-searchable.
Save the photo with the dish and it can be matched again from the camera.
-
This is early work.
Personal recognition grows with use. It is honest work in progress, not magic.
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.