Profile 👋

I’m a machine learning researcher working on reinforcement learning and language models.
I graduated with an MEng from ENPC and an MSc from the MVA M2 research program at ENS Paris-Saclay.

My research contributions include:

• distillation, pre-training at scale, Transformer architecture: Falcon, Apertus (ACL), FOG (NeurIPS)
• Reasoning, post-training: Terminator (ICML workshop, NeurIPS), Leanstral, Mistral Large 4.
• Algorithmic research on RL for long-horizon LLM agents at Mistral.

I joined Mistral AI’s science team in 2026 to work on long-horizon reinforcement learning, reporting to Albert Jiang.

Between early 2025 and early 2026, I was a visiting student at MLO lab at EPFL, supervised by Prof. Martin Jaggi. I was also a core member of the Swiss AI Initiative training team, an open-source initiative between EPFL, ETH Zurich, and CSCS.

Previously, I did an internship at the theory/frontier team of TII (UAE) and was a member of the Falcon LLM team.

🛎️ News 🛎️:

• [Nov 2026] ML4 (1T-A52B MoE) scientific report is published. It’s a good model!
• [July 2026] Weights + tech report + SafeVerify + FLTEval are now public, following Leanstral 1.5 release.
• [March 2026] Our paper on tackling overthinking in RLMs is finally out.
• [Nov 2025] Excited to present our poster on FOG at NeurIPS in Paris on November 25th.
• [Sept 2025] Apertus family of fully open LLMs is released!
• [Dec 2024] Falcon3 family of open models is out: distillation FTW!