 Ceramic waste forms are designed to immobilize radioactive materials for permanent disposal in geologic repositories. These waste forms must be compatible with the host material and resist degradation over long periods of time. Appetite and hollandite-structured ceramic waste forms have been studied recently, which demonstrate an approach combining computational techniques such as machine learning, first principles thermodynamics calculations, and kinetic-rate equation modeling to predict the composition of ceramic waste forms and their long-term dissolution rates. This approach has the potential to optimize ceramic waste form designs for problematic nuclear waste elements. This article was authored by Jianwei Wang, Dipta Beegosh and Zilong Zhang.