My Home AI Lab: Three Machines, No Cloud, ~200 Tokens/Sec
How I replaced cloud inference with a three-machine home stack — the hardware, the software, real benchmarks, and what a million tokens actually costs in electricity.
Building systems, writing about it.
I design and build infrastructure, security, and AI systems — then write up what I learned along the way.
Featured Projects
Overdrive
A hands-on AI inference lab for porting, quantizing, and benchmarking open-source LLMs across hardware — from a 12GB consumer GPU to multi-node setups. Focused on getting big models to run fast and cheap, with a documented experiment tracker for every result.
Agent Zero
An autonomous JSON AI agent framework — tool-driven task solving with persistent memory, skills, and plugins. The agent builds a system prompt, asks the model for one JSON tool request, executes it, and repeats until the task is done.