Alex Leung.

I'm a developer and independent researcher in the local LLM space.

I'm mostly interested in how small models behave on consumer hardware, especially inside coding agent harnesses. I look at where they loop or spiral, lose track of the task, or take tool output too literally, and how much of that a harness can actually change. Reika is what I use to measure it.

Outside of that, I tinker with inference engines, build small prototypes, and make quants and abliterations. Most of it stays exploratory, and some of it ends up here.