Understand your health risk — and exactly why.
One transparent engine, calibrated per disease. Enter a few routine health numbers and get a risk score and the precise contribution of every factor. Stroke and diabetes today — any condition tomorrow. No black box.
Explainable by design
Every prediction comes with an exact, per-factor breakdown — built on the model's own coefficients, not a guess. You see what raised and lowered the score.
Honest about imbalance
Stroke cases are rare (~5% of data), so raw accuracy is misleading. We optimize for ROC-AUC and recall with balanced class weights to actually catch at-risk cases.
Your data, your history
Optional accounts (Supabase auth) let you securely save assessments over time with row-level security — or use it instantly with no sign-up.
A transparent pipeline, end to end
From raw clinical data to an explained prediction in your browser.
Clean & encode
Median-impute BMI, keep 'unknown' smoking as a real category, scale and one-hot encode in one pipeline.
Train & select
Compare Logistic Regression, Random Forest and Gradient Boosting by cross-validated ROC-AUC.
Export to the edge
The winning model is exported to JSON and runs natively in the app — verified identical to scikit-learn.
Explain
Each factor's exact contribution is decomposed and shown as a diverging chart anyone can read.