We present BespokeMaterials, an innovative platform designed to streamline and accelerate workflows for materials design and engineering.

Using amorphous models of boron nitride and graphene as case studies, BespokeMaterials demonstrates how seamless integration with ab initio computational tools, such as Quantum ESPRESSO and SIESTA, together with AI agents, enables automated and highly efficient exploration and evaluation of physical properties of innovative advanced materials. Central to the described workflows is the linear scaling methodology, (www.lsquant.org)[1], which permits the computation of properties for atomic systems on an unprecedented scale—reaching billions of atoms. 

Our AI models are built on graph neural network architectures, due to their scalability for arbitrary input dimensions and their capacity to manage complex relationships in molecular structures. We leverage equivariant encoding and descriptors to encode the properties and interactions between atoms and molecules . This approach allows us to reduce the search space by taking advantage of spatial symmetries. By utilising AI models, we aim to decrease the time required to run a workflow without affecting its accuracy.

This framework represents a major leap forward in materials computation, enabling rapid development of AI-driven strategies for next-generation materials design and providing a comprehensive toolkit for both researchers and industry applications.