Someone has to read the label properly.
Indian shoppers are told to read the label and then handed one that cannot reasonably be read: INS codes with no names, panels declared per 100 g on packs that are 30 g, and claims on the front that the back does not support. Zabel is an attempt to close that gap without asking anyone to trust us on faith.
What we are building
A phone app that reads any Indian food label - barcoded or not - and turns it into something a person standing in an aisle can act on in a few seconds. It decodes the ingredient list, names every INS code and says what is actually known about it, converts the nutrition panel to the serving in front of you, and publishes a health score whose arithmetic is attached to it.
It works in English and in Hindi, and it works without an account. Scanning a pack, reading the result and reporting a mistake in ours all happen without signing in.
Who it is for
In order: parents buying for children under ten, adults managing diabetes, prediabetes, hypertension or PCOS, and people buying for protein and training. Those three pick the packs where getting the label wrong costs the most.
If you have told the app about a condition, that changes what a panel is read against and which line an explanation leads with. It does not change the score - and the profile itself never leaves your phone. See the privacy page for how that works.
How it is paid for
Not by the manufacturers. We take no money from brands, we sell no placement in the alternatives list, and we do not sell or broker personal data. A paid tier for people who use the app heavily is the intended way this supports itself.
That is not only a promise; parts of it are structural. The “better packs on this shelf” list is computed from the catalogue and never from who is asking, so there is no slot in it that could be sold even if someone wanted to sell one.
How we decide what a score means
The scoring method is versioned, and every reading records the version it was scored under. Thresholds are data rather than code, so a figure can be corrected without a release - and every change to them is a change somebody made on purpose and can be asked about.
Nothing a model wrote is published on its own. Extraction produces a candidate; a person promotes it. Every reading carries a “this is wrong” button that needs no account, and those land in a queue a human works through.
What we have not done yet
We would rather you heard this from us than found it out.
- No dietitian has co-authored the method. The arithmetic is published and defensible, but it has not yet been co-signed by someone with the standing to stand behind it. That is a blocker on launch, not a nice-to-have.
- The daily reference intakes have not been reviewed by a nutritionist. They are derived by a stated rule from ICMR-NIN and WHO figures. Stated rule, unreviewed numbers.
- The Hindi has not been reviewed by a native speaker. All of it is machine-written and is being reviewed before launch.
- The app is not on the Play Store. It is in development. The early access list is how you hear when that changes.
Where the data comes from
Part of the catalogue is built from Open Food Facts, a volunteer project, used under the Open Database Licence. The rest comes from labels photographed by people using the app and checked by us. Additive entries cite their sources, and an entry we have not verified says so rather than quietly reading as though we had.
Talk to us
Corrections are the most useful thing you can send, and they are welcome from anyone - you do not need to have used the app. The contact page reaches a person.