Tabular foundation models & NILM, EDF R&D (2024-present)

At EDF R&D, I work on tabular foundation models (TabICL, TabPFN, TabPFN-TS) for the energy sector, designing dedicated feature engineering for usage detection and consumption signal reconstruction. I also train, deploy and improve Transformer-based Deep Learning models for NILM (Non-Intrusive Load Monitoring), i.e. load curve disaggregation, and supervise internships on NILM, active learning and representation learning for time series (Mantis, Chronos, TimesURL).

Vélib' Métropole redistribution, [Smovengo](https://www.smovengo.fr/) (2023-2024)

Following our win at the Hackathon Vélib’ Métropole, I implemented the proposed method in production to optimize the morning redistribution of bikes across Vélib’ stations, combining optimal transport with a Deep Learning (MLP) model to arbitrate between distribution cost and bike availability.

Hackathon: [Vélib' métropole](https://blog.velib-metropole.fr/hackathon/) 2023

Competition: [challenge mathématiques et entreprise](https://challenge-maths.sciencesconf.org/) 2021

I’m also winner of the challenge mathématiques et entreprise, organized by AMIES. We worked remotely with the company Foyer (Leudelange, Luxembourg), on assessing and improving data quality using machine learning methods (interview).\