Applied ML, retrieval, and reliable production systems.
Professionally, I build machine learning systems from models to deployment. My work spans statistical learning, retrieval, ranking and evaluation, computer vision, numerical methods, and high-performance Python/C++ systems.
Independently, I maintain open-source numerical software and collaborate on retrieval and LLM systems, including search and RAG pipelines and tool-using applications. I also write about statistics, signal processing, and machine learning, especially where mathematical guarantees and observed behaviour diverge.
Applied mathematician by training, with a PhD in the field. Based in Paris.
Independent open-source work in applied mathematics and statistical computing.
robustcov — Robust covariance, scatter estimation, anomaly geometry, and contamination-resistant reference data for SHAP and LIME.
lattice-dsp — Stability-aware lattice methods for recursive digital filters.
cholrot — Rank-one Cholesky updates and downdates based on hyperbolic rotations.
Short notes on statistics, signal processing, and machine learning — mostly on where a method’s guarantees stop matching its behaviour.
Broader interests: statistical learning, kernel methods, time-series and state-space models, and the relationship between mathematical guarantees and observed behaviour.