Point, scan, decide — in three seconds
A barcode and label scanner that tells people whether a product fits their diet. I designed the scan-to-answer loop, the confidence and correction states, and the onboarding that teaches the app's limits honestly.
The problem
The recognition model is good, not perfect — and a nutrition app that confidently gives a wrong answer costs a user's trust permanently. The design had to be fast when certain and visibly careful when not.
The work
Designed the result screen around a single verdict people can read at arm's length in a supermarket aisle, with the reasoning one tap below it.
Created three distinct confidence states — confident, uncertain, unrecognised — each with its own visual language and its own next action.
Built a correction flow that takes seconds, so a wrong result becomes training data instead of an uninstall.
Wrote onboarding that states plainly what the scanner can and cannot read, setting expectations before the first failure rather than after it.
Optimised the camera screen for one-handed use with a trolley in the other hand — the actual context of use.
Key decisions
Every project has a handful of decisions that shaped everything after them. These were this project's.
A large, unambiguous yes/caution/no, then the numbers. Nobody reads a macro table while standing in an aisle.
When the model is unsure the UI says so and asks a single clarifying question, rather than guessing and being confidently wrong.
A two-tap correction path turns the app's worst moment into its best feedback loop.
Files were handed over annotated and component-based, with a live walkthrough for the engineering team.
More like this
Tell me what is not working. I will tell you honestly whether design is the right lever and what I would do in the first two weeks.
madnansakhi@gmail.com