We aim to describe the atomic and electronic structure of complex crystals at finite temperatures and pressure. The atomic/molecular constituents of these crystals show types of motions that cannot be described by simple harmonic springs, for example, rotations, rattling, or diffusing ions. We therefore construct inter-atomic potentials using sophisticated machine learning techniques based on the first principles electronic structure information. These potentials are hereafter used to simulate large crystal supercells under ambient conditions. We are especially enthusiastic about projects that combine (application-driven) method development with powerful experimental characterization techniques for promising novel materials for energy applications. Examples of such materials are perovskite solar cells, superionic conductors, and solid-state batteries.