Hello!

AtomDNN — ML framework for atomistic simulations

Open-source · containerized with Docker & Singularity · interfaced to LAMMPS

I am the author of AtomDNN, an open-source Python framework for training machine learning interatomic potentials (MLIPs) from density functional theory (DFT) data. The framework uses a modular architecture, automated testing, and reproducible workflows, and is directly interfaced to LAMMPS for large-scale molecular dynamics. It includes an adaptive loss-weighting algorithm to automatically balance competing objectives — energy, force, and stress prediction — in multi-component training.

Structural dynamics surrogate pipeline — 30,000× speedup

★ 30,000× runtime acceleration — finite element hours → milliseconds

During my Ph.D. internship at SLB (Schlumberger), I designed and deployed a full-stack surrogate modeling pipeline for real-time structural health monitoring of drilling components. The pipeline integrates automated Abaqus data extraction, proper orthogonal decomposition (POD) for dimensionality reduction, Gaussian Process Regression as the surrogate layer, and linear state-space models for time-domain reconstruction — reducing FEA runtimes from hours to milliseconds.

Fracture mechanics in 2D high-entropy VTiCrMoC₃ MXenes

I discovered the first-reported ductility and toughening mechanisms in 2D high-entropy VTiCrMoC₃ MXenes by combining MLIP-driven fracture simulations with statistical analysis of atomistic trajectories. Machine learning interatomic potentials trained on DFT data enable physics-accurate molecular dynamics at scales inaccessible to ab initio methods — spanning atomic to mesoscale.

Reversible phase transitions in monolayer MoTe₂

Using ML-accelerated multiscale simulations, I identified the kinetic mechanism for reversible stress-induced phase transitions in monolayer MoTe₂, focusing on the vacancy-assisted mechanism enabling this reversibility. Supported by NSF grant 2308163 and published in Communications Materials (2026).

Smart|DT — aircraft probabilistic damage tolerance software

I co-developed Smart|DT, an FAA-funded risk assessment software suite for aircraft probabilistic damage tolerance analysis at the Computational Reliability Laboratory (UTSA). I contributed to the Java-based GUI with modular architecture and automated testing, and accelerated probability-of-failure calculations from Monte Carlo methods to first and second order reliability methods (FORM/SORM). Smart|DT is interfaced to Afgrow, Nasgro, and Fastran codes.