Marvin Poul is a computational materials scientist working on machine-learned interatomic potentials, alloy thermodynamics, and computational methods combining them. His current focus is enabling back-propagation of costly thermodynamic simulation protocols for improved ML models of atomistic interactions.
He maintains and contributes to range of open source projects for computational materials science and related workflow tools, building automated workflows that take materials modeling from raw electronic structure calculations all the way to phase diagrams, and cares as much about the software craftsmanship as the physics it enables.