Universal Machine Learning Kohn–Sham Hamiltonian for Materials

I came to this paper looking for a way to move beyond using a MLIP model's latent space as a feature vector to represent a material in a computational inexpensive way.