BinaryBros · PeaceOfCode 2026
A rural health worker photographs a skin lesion. Tvak classifies the severity and likely condition in seconds — helping decide when to refer, without a specialist on-site.
The most important question for a health worker isn't what the condition is called — it's how urgent it is. Tvak is built around that. Severity classification is the primary output. Disease name is secondary.
Severity classification is reliable (86.5% test accuracy). Disease name prediction is less reliable on real-world smartphone photos — the model was trained on clinical images and may misidentify specific conditions while still correctly classifying severity. Always treat the severity level as the primary signal.