An evaluation framework for machine learning models simulating high-throughput materials discovery.

We test BERT on IS2RE task of Matbench Discovery.

Machine learning pipeline for next-generation battery materials discovery. Currently focused on predicting electronic band gaps using MatBench datasets and graph neural networks (GNNs). The goal is to extend the framework toward broader materials property prediction and generative design for advanced lithium- and sodium-ion battery compounds.

MatDesignKit: data-driven materials discovery pipeline using Materials Project/MatBench datasets plus OPTIMADE/NOMAD queries. Trains property predictors (matminer + CrabNet baselines), then performs multi-objective inverse design (Bayesian optimization) to propose synthesizable candidates matching target thermal/mechanical specs, with report + API.