Building small, fast, and
Kazakh

I like making things that actually run — models small enough to fit on a laptop, agents that don't pretend, pipelines that hold up when you ship them.

My working thesis: the next decade of AI will belong to people who can squeeze competent behaviour out of small, cheap, open-weight models — especially for languages and domains the frontier labs aren't optimising for. So I spend my nights on QLoRA runs, evals, agent scaffolds, and anything that lets one person punch above their weight.

Outside the terminal: bass-heavy playlists, long walks around Almaty, and a slowly growing collection of cassettes I can't actually play.