I am a PhD candidate at EPFL, advised by Prof. Pascal Frossard. My research focuses on building robust, efficient AI models with reasoning capabilities across different data modalities. I started by developing Meta-GNN, a personalized and highly efficient model for epilepsy detection specifically designed to operate on wearable devices. We then advanced to modeling causal reasoning capabilities by learning temporal causal structures from non-stationary data, resulting in Castor and Fantom frameworks. Currently, my work aims to make Large Reasoning Models more computationally efficient. Before joining EPFL, I completed my master’s degree at ENS Paris-Saclay (Mathematiques, Vision et Apprentissage: MVA), where I explored various aspects of AI and statistics. I also gained industry experience interning at Deezer Research, working on cross-domain recommendation.