Lead Software Engineer working where AI applications meet systems software.
I like problems with sharp edges: unsafe inputs, large datasets, hardware constraints, and software that has to explain why it reached a result.
Rust, C++, CUDA, and Python are tools. Making difficult systems measurable and dependable is the job.
| Project | What makes it interesting |
|---|---|
| kzip | Fast, memory-safe ZIP infrastructure for Rust and C, with async reads, encryption, differential verification, fuzzing, and a libzip-shaped ABI. |
| LodeStar | A C++ GNSS/SBAS integration, test, and verification platform spanning scenario generation, safety analysis, and certification-ready reporting. |
| AEGIS-PERC | An electrical-rule verification and semiconductor signoff platform with graph analysis, regression gating, and audit-ready workflows. |
| CSV-Arrow-Parser | A CUDA/C++ CSV-to-columnar parser with warp-level parsing and fused filter pushdown to avoid materializing discarded rows. |
| PatchOrchestrator | A multi-layer control plane for planning, supervising, pausing, and recovering fleet-wide software patch rollouts. |
| Healthcare Revenue Risk Predictor | A PyTorch system combining structured claims and clinical text to predict denial risk and missed billing opportunities. |
Understand → Model → Observe → Test → Ship → Measure → Refine ↺
The loop is deliberately circular: every release creates evidence, and that evidence sharpens the next problem definition. In practice, this means typed boundaries, deterministic tests, reproducible benchmarks, useful failure messages, and documentation that says what is not supported as clearly as what is.
Not every useful idea needs to become infrastructure. I also build experiments in mobile health, games, visualization, and developer utilities. You can find those in my repositories.
I'm building more of my work in public and welcome thoughtful collaboration. If a project overlaps with your work, open an issue with the use case, the failure mode, or the benchmark you care about. Specific problems make the best starting points.