The project
About Tvak
A triage tool for rural health workers — built to help decide when to refer, without a specialist on-site.
Mission
"In rural areas, a delayed referral for melanoma can mean the difference between life and death. Tvak puts triage capability in every health worker's pocket."
The problem
Why this matters
Rural health workers encounter skin conditions they can't identify. Without specialist access, they face a difficult choice — refer every patient to a distant hospital, or guess and send the patient home. For conditions like melanoma, that guess can cost someone their life.
Problem statement
"A rural health worker photographs a skin lesion or wound. Classify its severity or likely condition to aid referral decisions."
Performance
Model results
Tested on 3,674 images never seen during training. Severity accuracy is the primary metric — it's what drives the referral decision.
Limitations
What the model cannot do
Known limitations
Trained on clinical images — real-world smartphone photos under poor lighting may reduce accuracy.
Disease name is an estimate. Severity classification is more reliable and should be treated as the primary output.
Does not account for patient history, age, symptoms, or prior conditions — image only.
14 conditions covered. Many tropical and region-specific skin diseases are not in scope.
Not a medical device. Tvak is a proof-of-concept built for PeaceOfCode 2026. It should not replace professional medical diagnosis or treatment.