I'm a postdoctoral researcher in computational cosmology at the AstroParticule & Cosmologie Laboratory (APC, CNRS/IN2P3) in Paris, working on instrumental systematics, data analysis, map-making, and fast spherical-harmonic methods for the next generation of Cosmic Microwave Background experiments.
- Beam systematics — how an imperfect optical response leaks into the maps, and how to model it before it becomes a bias
- Map-making — turning time-ordered detector data into polarised sky maps, and keeping the operators linear enough to invert
- Spherical harmonics algorithms — fast, accurate transforms on HEALPix and other grids on the sphere
- Accelerated pipelines — JAX and GPU implementations of the above, differentiable where it helps
- Local LLMs — running open-weight models on my own hardware and building the tooling around them: tool-calling agent loops, quantised inference with
llama.cppand MLX, speculative decoding, and semantic memory over vector search - Occasionally other stuff — games in Godot and Pygames, news collections and analysis, predictive financial tools
| Project | What it does |
|---|---|
| FURAX | JAX-based framework for CMB component separation and map-making |
| tod_generation_mapbased_beam | Sample-based TOD generation: convolves polarised I/Q/U maps with a pixelated beam along the scan (docs) |
| HP2SPH_python | Fast, accurate HEALPix ↔ alm transforms via a double Fourier sphere, NUFFT and Slevinsky's FSHT — scalar and spin-2 |
| sparse_beam_matrix_creation | Builds a beam matrix at map level, skipping harmonic space entirely |
| LLM_tools | Tool-calling agent harness for locally hosted LLMs — file edits, code intelligence, git, search, memory |
| Project | What it does |
|---|---|
| Commander | Bayesian end-to-end CMB analysis by Gibbs sampling |
| s2fft | Differentiable, accelerated spherical transforms |
| jax-healpy | JAX implementation and extension of healpy |
Specialties: CMB Analysis | Beam Systematics | Map-Making | Spherical Harmonic Transforms | Automatic Differentiation | HPC | Local LLM Tooling
If the problem demands it, I'll learn the language.