Method Development & HPC Tooling
Workflow automation, MPI-enabled screening, and reproducible infrastructure for many-body calculations.
Motivation
Leadership-class simulations demand reproducible pipelines, data hygiene, and responsive visualization. I build tooling that bridges traditional electronic-structure packages with modern workflow engines so that large-scale screening stays transparent and repeatable.
Core Contributions
- MPI-aware task orchestration for WEST/BerkeleyGW calculations on Polaris, Perlmutter, and Frontier.
- Automated provenance capture and lightweight dashboards for monitoring calculation health.
- PyTorch-based surrogate models that plug into existing workflows to pre-screen candidate structures.
Resources
- Source and notebooks on GitHub
- Internal documentation (available to collaborators upon request)
Roadmap
Current tasks include integrating ML surrogates directly into WEST’s scheduling layer and publishing a reusable cookiecutter for multi-fidelity excited-state screening.