One battery, five decisions, five models.
A battery's value and risk come down to a handful of decisions: is this cell safe, when should the car charge, where should the next charger go, does this pack deserve a second life, and can we trust its history? Each project below takes on one of them with an open, working ML prototype. Together they form a small network that shares data along the lifecycle.
How these are built
Every model is compared with what people would do without ML: fixed thresholds, persistence forecasts, "test everything", "charge on plug-in".
Feature attributions, Shapley values and uncertainty bands are shown next to every prediction, so a decision can be challenged.
The data comes from physics-inspired simulators. Each page says what's simulated and what real data would be needed to validate it.
Plain Node.js with zero dependencies. The ML toolkit (random forests, isolation forests, ridge, k-means, kNN, dynamic programming) is written from scratch and runs on the server and in the browser. Printable QR cards →