Scratch Check
Point a phone at a scratch ticket and find out whether it won.
Scratch tickets are unpleasant to check. The rules are printed small, the symbols are inconsistent, and getting it wrong means throwing away a winner. Point a camera at the ticket and Scratch Check reads it and tells you.
It is the smallest product we run and the clearest example of a pattern clients ask for constantly: put a camera in front of something physical and get a reliable answer back. A model does the reading; when it is unavailable or the network is not there, on-device recognition takes over rather than the app simply failing.
That fallback is the interesting part. Anything that depends on a model needs an answer for the moment the model is not there.
What this one demonstrates
A camera-to-answer pipeline with a model doing the reading and a local fallback for when it cannot.
- Camera capture
- Model-backed reading
- Offline fallback