At Alstom, I designed a computer-vision solution for legacy engineering drawings that reached 89% detection and 85% classification accuracy in 8 weeks, using each drawing’s legend as guidance. I also led model analysis for predictive-maintenance anomaly detection, where the design outperformed the baseline.
At Institut Polytechnique de Paris, I researched human-motion analysis for physical rehabilitation, including robot control in simulation and on hardware. My PhysioFormer work set new state-of-the-art results on the KIMORE, UI-PRMD and KERAAL benchmarks; another framework advanced error classification and localization on KERAAL.
Earlier, I developed engineering-drawing algorithms at Spare Parts 3D and worked on C++ graphics drivers at NVIDIA. I’m completing a PhD in Deep Learning for Robotics, with my defence expected in December 2026, and have four first-author peer-reviewed publications.
