Granular materials are large amorphous aggregates of discrete, individually solid particles. Despite seemingly simple ingredients, such aggregates exhibit a wide range of complex behaviours that defy categorization as ordinary solids or liquids. This includes non-Newtonian flow behaviour and collective 'jamming' transitions. One of the key issues has long been how to link particle-level properties in a predictive manner to the behaviour of the aggregate as a whole. However, for actually designing a granular material, an inverse problem needs to be solved: for a given desired overall response, the task becomes finding the appropriate particle-level properties. This master class discusses new approaches to tackle the inverse problem by bringing concepts from artificial evolution to materials design. These results have general applicability and open up wide-ranging opportunities for materials optimization and discovery.