Machine learning interatomic potentials (MLIPs) allow quantum-mechanical accuracy to be achieved at a fraction of the cost of direct DFT calculations, enabling simulations at experimentally relevant timescales and system sizes. My work focuses on training and fine-tuning equivariant GNN potentials — in particular the MACE architecture — for complex inorganic and molecular systems. A recent highlight is the fine-tuning of MACE-MH-1 on targeted RuP datasets, enabling the discovery of a two-stage phase transition that was inaccessible to direct DFT-based MD. I also develop and maintain multi-GPU LAMMPS+MACE workflows for LUMI (AMD ROCm) and Mahti (NVIDIA).

NMR spectroscopy is one of the most powerful probes of local structure and dynamics in materials, but first-principles computation of chemical shifts is expensive, limiting achievable statistical sampling. I develop ML NMR models — trained on relativistic DFT shielding tensors — that predict ¹²⁹Xe, ¹³C, ¹⁵N, and other nuclear chemical shifts at negligible cost once trained. A key focus is xenon NMR in porous organic cages (CC3) and porous liquids, where cage breathing, solvent reorganization, and guest siting collectively determine the observed spectrum. Models are deployed on-the-fly during MD trajectories, enabling direct comparison with variable-temperature experimental NMR data.

Classical molecular dynamics treats nuclei as point particles, neglecting quantum effects such as zero-point energy and tunneling that are particularly important for light nuclei and at low temperatures. Path integral molecular dynamics (PIMD) addresses this by representing each nucleus as a ring polymer of classical beads, recovering quantum statistical mechanics exactly in the limit of many beads. My work combines PIMD with ML interatomic potentials to make quantum nuclear simulations tractable for large systems. A key application is C₆₀ fullerene, where quantum effects significantly influence both structural properties and ¹³C NMR chemical shifts, enabling direct comparison with experimental solid-state NMR data.

Deploying ML potentials at scale requires careful integration with HPC infrastructure. I develop and document production-ready workflows for running LAMMPS with MACE on both AMD ROCm (LUMI) and NVIDIA CUDA (Mahti) architectures, resolving compatibility issues between ML libraries, MPI implementations, and GPU backends. These workflows — including SLURM job scripts, module environments, and build instructions — are made publicly available. I also maintain automated VASP DFT pipelines for generating ML training datasets, and contribute Python analysis tools for RDF, structure factors, space-group tracking, and phonon dispersion from MD trajectories.