-
I just like coding and solving problems. Currently working as MLOps Engineer for a supplier of post-trading services.
- Joined on
2025-05-28
Adrien Bouvais
Data, Platform & MLOps Engineer based in Luxembourg.
I design backend systems, data platforms and inference infrastructure, currently on the backend of Clearstream AI at Clavem Group. Outside of work I write Zig, mostly tools for physics and simulation, dev video games and self-host my own server at home.
Currently building
Zigma: a programming language for scientific computing
Zigma makes scientific computing safer and easier. It attaches dimensions and scales directly to variables and performs strict dimensional analysis at compile time: add meters to seconds and your code simply won't compile. The source isn't public yet, but the documentation is.
g = 9.81 m/s2
t = 0..10 s step 0.1
v0 = 20.0 m/s
// Zigma automatically derives that 'y' is in [m]
y = (v0 * t) - (0.5 * g * t^2)
// Compile-Time Error: Cannot add [m] to [s]
wrong = y + t
- Math-first syntax: reads like a math notebook
- Dimensions first: attach dimensions to variables and let the compiler check them
- Automatic scale: add meters to inches and Zigma does the conversion
- Python and Jupyter: designed to be used in notebooks alongside Python and NumPy
- Fast: built on Zig, with SIMD or WebGPU
- Uncertainty propagation: confidence intervals (
20.0 +/- 0.5 m/s) handled natively in math operations
dimal: dimensional analysis for Zig
The type system powering Zigma. A single Tensor API parameterized entirely at compile time by:
T (numeric type) · dims (physical dimensions) · scales (unit scale) · shape (array shape)
Originally built to store i128 positions in space simulations and avoid float precision loss at astronomical scales. Its WebGPU binding lives in zig-wgpu.
Trading game
A market simulation built as one giant DAG: nodes generate, transform, store, trade and consume resources, all with dynamic pricing. You start small, buying and selling tiny quantities, then grow until you're moving the market itself. Written in Zig + Raylib, no modern engine.
Work
Senior Data/ML Engineer, Clavem Group · 2024 – present Backend and architecture of a cloud-agnostic AI/data platform used in production by multiple business units.
- Design and upkeep of the platform architecture: ingestion, storage, serving and security
- GitOps CI/CD on OpenShift + ArgoCD, with a config repo per environment and PR-gated promotion to production
- Versioned model and dataset registry (DVC, Artifactory, Airflow) so Data Science teams work autonomously
- Least-privilege access, secrets management and compliance checks before every release
Data Engineer, Reveals SA · 2023 – 2024 Core banking data migration and quality platform for a Luxembourg bank.
- Fault-tolerant pipeline migrating billions of sensitive rows: zero data loss, encrypted in transit
- Airflow DAGs for multi-source ETL (Oracle → PostgreSQL), plus pseudonymisation and synthetic test data
- Real-time data quality platform (Flask + HTMX) and a self-service SQL validation framework for non-technical users
Data Engineering Consultant, Ernst & Young · 2022 – 2023 Data architecture advisory for enterprise clients on Azure: platform design, SQL schema optimisation, Airflow-orchestrated ETL.
Education: Engineering degree in Thermal & Energy Engineering, Université Bretagne Sud (2019 – 2021)
Other projects
| Repository | Language | What it is |
|---|---|---|
| ZipponDB | Zig | A database written in Zig |
| zig-multi-thread-raytracer | Zig | A multi-threaded raytracer |
| Advent-of-Code | Zig | Puzzle solutions |
| Selfhosted-config | CSS | Docker Compose + config for my home server |
| nvim-zig | Lua | My minimal Neovim config for Zig |
| Dotfiles | Lua | Arch Linux setup |
Stack
| Languages | Python Go Zig SQL Shell |
| Data | PostgreSQL Oracle SQLite Airflow DVC |
| Platform | OpenShift ArgoCD Terraform Docker GCP Azure |
| CI/CD & APIs | GitHub Actions FastAPI Flask |