Interpretable machine learning / quantum mechanical platform for accelerated materials discovery

This listing presents a physics-informed and interpretable machine-learning platform for accelerated materials discovery, integrating first-principles simulations (DFT), advanced descriptors, and Kolmogorov–Arnold Networks (KANs) to predict electronic, thermodynamic, and transport properties of complex materials. The framework has been validated on diverse datasets including bulk crystals, low-dimensional materials, and interface systems, achieving competitive accuracy (R² > 0.85 for key targets such as band gap and Seebeck coefficient) while preserving interpretability through symbolic and feature-resolved representations.

The platform enables rapid virtual screening, inverse design, and mechanistic insight for applications in energy materials, catalysis, photonics, and electronic interfaces. Current proof-of-concept studies include high-throughput screening of thermoelectric compounds, defect-engineered semiconductors, and spectroscopy-driven molecular identification workflows. Ongoing work focuses on validation with experimental partners, uncertainty-aware prediction, and deployment in pilot industrial use cases. We are seeking partners for co-development, pilot testing on industrial datasets, and joint translation toward application-ready decision-support tools for materials design.