The project aims to integrate artificial intelligence, high-performance computing, and computational materials science to accelerate the discovery of novel functional materials, with a particular focus on magnetic systems. As a key component connecting broader research efforts in data-driven materials design and computational magnetism, the project centers on the development of advanced machine-learning methods and scalable simulation tools for materials informatics. A major objective of the current phase is the development of a novel software framework for atomistic spin dynamics, implemented in JAX and designed from the outset for efficient GPU execution. By combining machine learning with large-scale magnetic simulations, the project seeks both to identify promising magnetic material candidates and to establish a high-performance computational framework for their investigation. Scientific and Technical Objectives: The proposed work has two closely connected goals. First, we will develop and apply machine-learning and data-mining workflows for screening and identifying candidate magnetic materials from large materials datasets. These workflows will support prediction, filtering, and prioritization of compounds with desirable magnetic properties. Second, we will design, implement, test, and optimize a GPU-native atomistic spin dynamics software package written in JAX. In contrast to conventional CPU-oriented codes, this framework is being developed specifically to exploit modern GPU architectures. This enables efficient large-scale parallelization for simulations of magnetic dynamics, parameter sweeps, and exploration of complex magnetic configuration spaces.