AI & blockchain developer based in Montpellier. I explore language models, from training them to running them locally. Into Rust and open-source tools, and understanding what happens under the hood.
Selected projects
01 / 05A little closer to the metal
02 / 05Large language models.Right on your Mac.
MLXL3’s direct-Metal backend manages GPU buffers, command queues and compute kernels from Rust. EXL3 model execution runs on Apple Silicon without an MLX runtime in this backend.
The direct-metal backend dispatches Metal kernels on Apple Silicon without the MLX runtime.
03 / 04Things I like working on
03 / 05Local inference
From quantized weights to token streams. Running models locally, understanding memory, and making the most of Apple Silicon.
Rust & systems
Native runtimes, low-level implementation, and tools with clear boundaries. Close enough to the machine to understand what it is doing.
Blockchain
Smart contracts, cryptography and the infrastructure behind decentralized applications. Exploring how to encode trust in a system.
Models & agents
Training, behavior, evaluation and developer tools. Experiments that turn an interesting question into something you can inspect.
A bit about me
04 / 05Building things.Taking them apart.
I’m Justin, a developer based in Montpellier. I have a soft spot for Rust and systems you can understand, run and make your own.
Blockchain & cryptography
Smart contracts and the foundations of trust. EventChain brings these ideas into a team-built platform that encourages people to show up.
Language models
Understanding architectures, training and behavior. frlm explores French language models; Heretic NX focuses on behavioral editing and evaluation.
Local intelligence
From runtime to interface, MLXL3 and ClawRS explore local model execution. With locked, I’m looking at tool isolation and keeping agent actions traceable.
An open notebook, not a black box.
Research notes, tools and unfinished ideas. Always experimenting.
Explore the experimentsGet in touch
05 / 05Good work starts witha good conversation.
An interesting problem, a question about a project, or just comparing notes. You can find me here.
Thanks for stopping by.
À bientôt.