Data-driven methods offer powerful new ways to navigate the coupled composition–processing–structure–property relationships that govern structural materials. In this seminar, I will illustrate this potential through two applications: designing novel multi-principal element alloys (MPEAs) and extracting insights on their deformation behavior from atomistic simulations. First, I will present recent efforts to optimize the mechanical properties of MPEAs. Most data-driven studies of high-entropy alloy design have focused primarily on composition, often within restricted elemental spaces and using simulated proxy metrics. To move beyond these limitations, we developed a machine-learning-based, surrogate-assisted, multi-objective optimization framework that jointly explores composition and processing in a design space spanning 15 candidate elements and 5 processing parameters. The resulting surrogate models capture physically meaningful processing–property relationships, both across the overall design space and within selected compositional subspaces. I will also briefly highlight complementary efforts to improve materials-focused Large-Language Models (LLMs) by constructing smoother latent-space representations that preserve the semantic structure of materials descriptions. On the modeling side, I will present results on data-driven analysis of atomistic simulations in a model refractory MPEA using a symmetry-adapted basis, namely Strain Functional Descriptors (SFDs). The SFDs are derived using Clebsch-Gordan coupling from invariants of nth order central moments of local number density, based on a Gaussian kernel. SFDs characterize shape, size, and orientation of local atomic environments, can be mapped to second and higher order deformation metrics as defined in continuum mechanics, and represent a minimal and complete basis for quantifying deformation. In conjunction with Convolutional Neural Networks (CNNs), ensemble averaged SFDs can forecast