Summary
Josselin Roberts is an AI Scientist based in Palo Alto with five years of experience bridging cutting-edge research and production software. He has contributed to foundation-model evaluation and high-throughput reinforcement learning at Stanford, applied world-modeling objectives in a FAIR video-modeling internship, and now works on next-generation multimodal models at Mistral AI. His background spans perception engineering at NVIDIA and optimization/heuristics for vehicle routing at Google, giving him a rare blend of systems, ML, and applied research expertise. Josselin is comfortable across the stack—from full-stack platforms and ETL pipelines to training large models—and has a track record of scaling algorithms to huge batch sizes enabled by novel research engines. Trained at École Polytechnique and Stanford, he pairs rigorous mathematical foundations with pragmatic engineering, and often gravitates to projects that translate research prototypes into robust, high-throughput systems. A lesser-known thread in his career is repeated work on equitable human-centered platforms (e.g., Stanford Deliberation Platform), reflecting attention to real-world impact beyond model metrics.
5 years of coding experience
3 years of employment as a software developer
Engineering Degree Computer Science and Applied Mathematics, Engineering Degree Computer Science and Applied Mathematics at École Polytechnique
PTSI/PT* Mathematics Engineering Physics, PTSI/PT* Mathematics Engineering Physics at Lycée Jean-Baptiste Say, PTSI/PT*
Master's degree Computer Science, Master's degree Computer Science at Stanford University
French, English, Spanish