Atoms are the Lego blocks of the universe. Stack them differently, and you get everything from diamond to carbon dioxide. Chemists used to tinker for months in the lab to test a single combination. Now, the MatGL program, like a virtual construction kit, instantly predicts the properties of any assembly. You upload an atomic blueprint and find out whether the material will be transparent or super-strong. Even how it will behave under a beam of light (spectroscopy). For instance, plain carbon can become graphite or a nanotube.

The secret of MatGL is graph neural networks: algorithms that see atoms as points and the bonds between them as lines. The same technology that suggests friends on social media here seeks out perfect atomic pairs. Trained on millions of examples, the network flawlessly guesses properties, even for tiny hydrogen in different compounds.

The library is free and open — a digital sandbox for creating materials of the future: from batteries to water filters. Months of waiting have turned into hours of computation. A scientific Lego set where the rules of physics work faster than intuition.