Next.js / FastAPI / MobileNetV2 / OpenCV
ScanMint
Awaiting Footage
Photograph the paper, ScanMint figures out what to do with it.
ScanMint finds documents inside imperfect photographs, straightens and cleans them, then learns what it’s looking at — whether that’s an invoice, form, receipt, or handwritten note. What started as a computer-vision experiment became an attempt to make paperwork feel a little less like paperwork.
Placeholder // divider
A prediction isn’t the end of the pipeline.
When ScanMint gets a classification wrong, the user can correct it. Those corrections become labelled examples for future training, turning ordinary use into a growing source of feedback. The model isn’t treated as finished; it’s designed to get better at the thing it keeps getting wrong.
