Draw any digit from 0 to 9 in the box with your mouse, trackpad, or finger. The network predicts what you wrote automatically and shows a confidence score for every digit.
Your drawing is normalized — cropped to the ink, recentered, and scaled — then scored against a neural network trained on handwritten digits. The output is a probability for each digit 0–9.
A small neural network trained on the MNIST dataset of 60,000 handwritten digits — the same family of models that first taught machines to read handwritten mail.
Around 95–97% on held-out handwritten digits, with each prediction returned in well under 50 milliseconds.