Current
PhysicsNeMo
NVIDIA · 2024 – present
Open-source framework for AI in physics and engineering (3k+ GitHub stars, 100+ contributors). I’m the top contributor over the past year and the core architect of the repository, including domain parallelism, the datapipes, and several end-to-end training recipes.
GeoTransolver
NVIDIA · Co-author, with R. Ranade
PhysicsNeMo’s top-performing surrogate model for CFD and CAE, using multi-scale, geometry-aware physics attention on irregular domains.
Past
CosmicTagger
Argonne · Lead developer
Deep-learning image segmentation (UResNet) that separates neutrino interactions from cosmic-ray background in neutrino detectors. Beat the previous state of the art by 5x in accuracy, and became a benchmark and acceptance test on several supercomputers. PyTorch, JAX and TensorFlow versions, scaled to thousands of ranks.
QMC
Argonne · Lead developer
Differentiable solver for the many-body Schrödinger equation in JAX, using neural-network wave functions for light nuclei. Scales with mpi4jax to thousands of ranks.
larcv3
Argonne / DeepLearnPhysics · Lead developer
C/C++ data serialization for sparse and irregular data, built on HDF5 with a Python interface. Scaled to 20k+ MPI ranks.
SparseEventID
Argonne · Lead developer
Sparse 3D computer vision models for particle detectors: classification, object detection, and SimCLR self-supervised representation learning. PyTorch.
Event displays for neutrino experiments
Yale / Harvard · Author
PyQt and pyqtgraph visualization tools for interactive viewing of detector data, used across several neutrino experiments and the DeepLearnPhysics group.