I have over a decade of experience in professional software dev, and an additional decade prior to that as a hobbyist. I've run my own business, worked as a freelancer, and worked as an employee in startups in every funding stage from Seed to Series C.
In my time outside of work, I'm building local AI inference stacks.
-
I wrote Titanic 4.1 from scratch in Rust as the fastest single-stream inference stack for DeepSeek V4.1 Flash's native weights across 4x DGX Sparks.
-
I wrote GLM 5.3 Flash Rust from scratch (again, in Rust) as a balanced quality/speed inference stack, skewing toward quality, for 4x DGX Sparks.
-
Currently working on other experiments as well (at the time of this writing, exploring how best to leverage an RTX 5090 in a Thunderbolt 5 eGPU to accelerate Qwen 3.8 Flash Next on a 128GB M5 Max MacBook Pro).