BeyondLabel

Packaged food in India sells on front-of-pack claims: high protein, no refined sugar, natural. Labelling rules are weak, and the fine print is written for nutritionists.

My first idea was the obvious one: point your phone at a label and get a simpler version of it.

Talking it through with friends and family, the label wasn't what tripped people up. They fell into two groups. Some thought a food was healthy when it wasn't: digestive biscuits, muesli, protein bars. Others knew chips were unhealthy but had no feel for how unhealthy.

A cleaner label fixes neither. Both groups needed the consequence, not the data.

How strong is this? Informal, untracked conversations. Enough to choose a direction, not enough to prove it, which is why the success metric below is the one that would show me I'm wrong fastest.

Scan or upload a label and pick a goal: general health, muscle gain, fat loss or managing blood sugar. BeyondLabel returns a plain verdict, explains what the product does to your body for that goal, and calls out tricks common on Indian labels, like soy passed off as premium protein, jaggery sold as "no refined sugar" and shrunken serving sizes. When a product fails, it suggests a swap that satisfies the same craving.

• Live barcode scannerI built it, then removed it. The live camera feed failed inside Instagram and WhatsApp's in-app browsers and crashed on multi-lens Android phones. A scanner that fails some of the time teaches people not to trust it, so the app now hands the camera to the phone's own picker.
• WhatsApp bot versionThe lowest-friction option of all, but Meta's API costs grow with every message, and a chat thread can't hold the history and streaks that bring people back.
• Calorie loggingLogging every gram is the chore that makes people quit tracking apps. BeyondLabel asks for one photo, not a diary.
• Choosing a goal before the first scanNew users start on general health, so the first verdict arrives within seconds. Goals live in the profile for people who want them.

• Images are compressed to 800px and run on a lighter model so verdicts fit inside the hosting time limit. I gave up some image detail for an app that doesn't time out.
• There are no accounts at first. History lives on the device, so trying it costs nothing.

Run the swap metric with 20 people for two weeks. If fewer than 1 in 10 scans lead to a swap, the problem isn't awareness either, and the product needs rethinking, not polishing.