From Panel Data to Token Sequences
NLP-inspired Machine Learning for Longitudinal Categorical Data
May 12, 2026
About
I’m a French civil servant of the Statistician-Economist corps, currently working as a data scientist at the AI and Data Science lab of Insee, France’s national statistical institute. My open-source projects include torchTextClassifiers (a PyTorch toolkit for text classification), MLOps pipeline design for automatic NACE coding, satellite imagery analysis for census monitoring, and an agentic framework for hierarchical nomenclature reasoning. I also contribute to AIML4OS, a EU initiative fostering AI/ML adoption across National Statistical Institutes, notably through an MLOps tutorial.
Alongside this, I’m a PhD candidate at Crest (ENSAE/CNRS) and Inria SODA, in the Graph4Health project - advised by Guillaume Lecué and Gaël Varoquaux. I’m working on Transformer-based representations of medical pathways for causal inference and policy evaluation using the French national healthcare database.