Tinkerer. LLMs, local models, and the occasional rabbit hole.
macring — distributed LLM inference across Macs over a Thunderbolt ring, built on Apple MLX. Prefill and decode run on different machines in the ring, which lets you fit bigger models than any single Mac can hold.
mlx-bench — small benchmark harness for MLX models: tokens/s, peak memory, and cold-start time across a set of standard prompt shapes.
llm-tap — tiny CLI for chatting with local models straight from the terminal, no server required.
dotfiles — my Linux setup: shell, editor, and the odd script.
Ask me about any of my projects — I'm happy to chat!