ML/AI · 2025
DermAI Demo
On-device dermatology decision support for low-resource clinics. A quantized deep learning classifier that runs entirely on the phone.
The problem
Dermatology expertise is scarce outside major cities, and connectivity in rural clinics is unreliable. Inference should never wait on a server round trip.
Approach
- Trained a classifier for common skin conditions, then quantized to INT8 TFLite for mid-range Android hardware.
- Native C / ARM NEON preprocessing bridged into Flutter through Dart FFI.
- Server-side Grad-CAM and SHAP explainability when a connection is available.
Why it matters
The whole inference path works offline and degrades gracefully where bandwidth does not exist.