The "hello world" of machine learning, made tangible.
Explore the classic handwritten-digits dataset, train a model to recognise digits, then test its accuracy yourself — machine learning's most famous exercise, running live in your browser.
React · Docker
This tool is temporarily offline for maintenance — check back soon, or get in touch if it's urgent.
Every machine-learning practitioner starts with MNIST — a dataset of 70,000 handwritten digits that has served as the field’s standard first exercise for decades. The MNIST Demystifier puts that exercise in your browser: look at the data, train a recogniser, and interrogate what it learned.
No prior technical knowledge assumed. If you’ve ever wanted the phrase “training a model” to mean something concrete, start here.
Browse the actual handwritten digits models learn from — seeing the data is half of understanding the problem.
Kick off training yourself and then probe the trained model's accuracy — including where and why it gets digits wrong.
MNIST Demystifier runs live — no sign-up needed to look around.
Open MNIST Demystifier→This tool is temporarily offline for maintenance — check back soon, or get in touch if it's urgent.
