Incoming researcher at MILeS (CNRS, Dauphine-PSL), continuing the work on LLM reasoning from my internship there.
I'm also wrapped up in recursive self-improvement, emergence and AI safety.
Recent Research
↗Selected Projects
↗Agent Ground Up
2026Coding agent from scratch: constant-size memory, recursive self-improvement, hand-derived DAPO loss.
Latent Intent
2026AI-safety interpretability work toward detecting hidden objectives and scheming, even when an LLM's outward behavior is held fixed.
Trajelysis
2026Framework for visualizing, decomposing, and analyzing layer-wise latent trajectories during generation.
Agentica
2024Modular agentic framework for automation, learning, and research.
