Bio
I’m a PhD candidate in computational cosmology at the AstroParticule & Cosmology Laboratory (APC, CNRS/IN2P3) in Paris. I build differentiable, distributed cosmological simulations for full-field weak-lensing inference, with an eye toward LSST to constrain cosmological parameters.
I’m also part of the Simons Observatory, working on CMB component separation. I’m particularly interested in modeling the spatial variability of foreground spectral indices and using the FURAX framework to reduce bias on the tensor-to-scalar ratio r in full-sky analyses.
I am also interested in combining CMB and large-scale structure datasets for cross-survey analyses to tighten constraints and improve systematics control.
My work sits at the interface of cosmological modeling and high-performance computing—distributing workloads with JAX and NCCL, and using modern automatic differentiation to build end-to-end differentiable models. Tools I lean on include C++ and CUDA.
Before CNRS, I worked in industry at Dassault Systèmes as an infrastructure engineer on CATIA and SOLIDWORKS; earlier, I was a data acquisition engineer at Renault, focusing on ADAS sensor data collection and pipelines for algorithm development.
Projects & Packages
- jaxDecomp — Distributed 3D FFTs and multi-GPU domain decomposition on large 3D grids in JAX.
- JaxPM — Particle–mesh cosmological simulation toolkit in JAX (Distributed).
- FURAX — “Framework for Unified and Robust data Analysis with JAX” for inverse problems in astro/cosmo.
- jax-healpy — JAX-native HEALPix/Healpy utilities, GPU-ready and autodiff-aware.
- jax-grid-search — Distributed grid search + gradient-based optimization built on JAX/Optax. (Also on PyPI.)
Contributions
- S2FFT — Differentiable spherical/Wigner transforms (JAX & PyTorch).
My contribution: translated part of the spherical harmonics algorithm to CUDA to avoid long JAX JIT compile times.