Sebastien Dubois
Engineer focused on performance — CUDA, AI toolchains, physics simulations.
I enjoy shipping fast, reliable systems: profiling and tuning kernels, crafting GPU pipelines, wiring AI workflows, and keeping simulation code practical and maintainable.
- Years shipping
- 3
- Benchmarks run
- 800+
- Institutions
- 3
From research labs to production rollouts
MPI, CUDA, and Slurm workloads profiled
CMAP · ONERA · CEA
Selected work
Projects
AutoPerf — HPC benchmarking & optimization platform
Experience & education
CV
Timeline and export
Latest posts
Blog
Notes on CUDA, toolchains, and HPC
Activity & languages
GitHub
Operating principles
- Reproducible performance first: changes are gated by tests and benchmarks (ReFrame + Google Benchmark).
- Ship maintainable systems: small APIs, modular C++/Python, design docs, and reviewable diffs.
- Automate the loop: CI/CD with GitHub Actions, orchestrated runs (Slurm/Redis), and observable runbooks (Sentry).
- Partner with scientists: turn research code into reliable components with clear interfaces.
Stack snapshot
- LanguagesPython · C++ · Fortran
- HPCCUDA · MPI · OpenMP · Kokkos
- OrchestrationSlurm · Docker · Redis
- DataPostgres · S3 · Vector DBs
- CIpytest · ReFrame · GitHub Actions