Projects & Research
A collection of research models, scientific tools, and code implementations.
DSpinGNN
[GitHub]An E(3)-equivariant Graph Neural Network combined with a Residual MLP architecture designed to predict magnetic exchange coupling parameters (J) in materials like CrI3. Successfully scales first-principles calculations from 8 atoms to large-scale simulations containing 3,200 atoms.
SpinDFT
[GitHub]An automated, crash-proof HPC orchestrator integrating Quantum ESPRESSO, Wannier90, and TB2J for high-throughput data generation.
Phystrackx
[GitHub]An independent, proprietary scientific tracking suite engineered for experimental physical laboratories. Streamlines sub-millisecond video tracking, real-time trajectory visualization, and high-frequency data acquisition.
NequIP (From Scratch)
[GitHub]A clean, modular reproduction of NequIP (E(3)-equivariant neural network for interatomic potentials) developed completely from scratch using PyTorch and the e3nn library. Built to deeply understand equivariant linear layers (o3 Linear) and tensor products.