Metallic thermoelectric alloys are promising for active cooling, where high thermal conductivity and a large power factor together maximize heat transfer, a design criterion distinct from conventional Peltier cooling. This space remains underexplored, with no tailored machine-learning framework reported. We build a curated experimental database of binary metallic alloys and introduce a hierarchical framework for Seebeck coefficient prediction, using a three-tier progression, global regression, sign-classified regression, and sign- and temperature-stratified regression, to mitigate dataset imbalance and capture the asymmetric, temperature-dependent behavior of metallic systems. This strategy improves transferability to unseen alloys, validated experimentally in Ni-Co, Al-Ni, Cr-Mn, and Ni-Cr. Feature-importance and SHAP analyses identify DOS asymmetry near the Fermi level, electronegativity deviation, and magnetization as the key descriptors, consistent with established thermopower physics. Several unexplored candidate alloys are also identified, establishing an efficient and interpretable route to metallic thermoelectrics for next-generation active cooling.

Solid-state thermomagnetic modules operating based on the Nernst–Ettingshausen effects are an alternative to conventional solid-state thermoelectric modules. These modules are appropriate for low-temperature applications where the thermoelectric modules are not efficient. Here, we briefly discuss the application, performance, similarities, and differences of thermoelectric and thermomagnetic materials and modules. We review thermomagnetic module design, Nernst coefficient measurement techniques, and theoretical advances, emphasizing the Nernst effect and factors influencing its response in semimetals such as carrier compensation, Fermi surface, mobility, phonon drag, and Berry curvature. The main objective is to summarize the materials design criteria to achieve high thermomagnetic performance to accelerate thermomagnetic materials discovery.