Hands-on AI projects, built end to end
Each project here goes from public dataset to trained model to a working app — with the real numbers, the dead ends, and the export gotchas included. Everything is educational: built to learn and to teach, not to be a product.
Educational content only. Nothing on this site is medical, financial,
or professional advice, and none of these projects are certified products or devices.
See the full disclaimer.
Projects
MoleCheck
A skin-lesion risk-awareness app: YOLO11 classifier trained on 11,720 dermatology images, running fully on-device in a Flutter app. No photo ever leaves the phone.
ECG Arrhythmia
A 1D CNN that classifies heartbeats from the MIT-BIH database — and an honest look at the inter-patient generalization gap, where 96% in the lab becomes 44% on unseen patients.
Next project
In progress — a new end-to-end build is underway. Write-up lands here when it ships.