PhD Candidate in AI · École Polytechnique · LIX
Palaiseau · France
I work on representation learning, world models, and prognostics for complex systems, at the intersection of deep learning and operations research. My PhD is part of the ORAILIX group at LIX, with IRT SystemX and Safran Tech, on probabilistic estimation of temporal representations in degrading systems. Before the PhD I cofounded Autoderm (medical AI for dermatology) and built Rust infrastructure for inference at the edge.
École Polytechnique, LIX, ORAILIX group · Palaiseau, France
Thesis: Probabilistic Estimation of Temporal Representations in Complex Systems. Supervised by Prof. Jesse Read, in collaboration with IRT SystemX and Safran Tech under the JNI3 program. Focus: representation learning, uncertainty quantification, world models, health indicators, prognostics under realistic degradation and maintenance.
DACS, Universiteit Maastricht · Netherlands
Master thesis: Navigating WebAI, training agents to complete web tasks with large language models and reinforcement learning. Coursework in deep learning, multi-agent systems, NLP, and knowledge representation.
Universiteit Maastricht · Netherlands
Palaiseau, France
Barcelona, Spain
Santa Cruz de Tenerife, Spain · Hybrid
Full list on Google Scholar. Several follow-up papers using TurboSens are under review at other venues.
Paper presentation, Porto, Portugal.
Paper presentation on uncertainty quantification for turbofan engines.
Poster, Trocadéro, Paris.
CentraleSupélec, France.
Ávila, Spain.
IoT Evolution Product of the Year award by TCM.
Guest lecturer.
École Polytechnique
Deep learning methods for complex system degradation modeling.
École Polytechnique
Deep learning methods for complex system degradation modeling.
With team Piaget: Wiring the Fly Brain into a Hierarchical World Model. Under the patronage of Philippe Baptiste and sponsored by Yann LeCun on the JEPA architecture.
For The Last Transmission, on the theme “Intelligences?”. Announced 20 May at the Panthéon, Paris.
With Maastricht University: applied hybrid-quantum solution in generative design, culminating in a 3D-printed bike structure designed on IBM QX.
Three national records, two international medals; 4th at the 2017 World Championship in Tomsk, Siberia.
A turbofan substrate for representation learning, prognostics, and reinforcement learning, with paired ground-truth latent state. Used in the ICML 2026 RLxF paper; multiple follow-up papers under review.
JEPA world model trained on spikes from the Drosophila connectome (FlyWire), embodied in NeuroMechFly v2 inside MuJoCo. Coupled along the connectome’s own sensory → integration → motor flow.
Training language-model agents to complete web tasks via reinforcement learning, in a custom simulated browser environment.
Combatting overfitting in classical decision tree classifiers with quantum-assisted post-pruning.
Available on request. Primary references: Prof. Jesse Read (LIX, PhD advisor), Prof. Sonia Vanier (ORAILIX), Dr. Rim Kaddah (IRT SystemX), Dr. Guillaume Doquet (Safran Tech), Dr. Mirela Popa and Dr. Gerasimos “Jerry” Spanakis (Maastricht).