I'm Mike McMahon — infrastructure/SRE engineer, AI practitioner, homelab builder. This site is where I document the work: what I'm building, how it broke, and what I learned along the way.
Use the sidebar to navigate. Start with How I Got Here if you're new, or jump straight into the Session Notes for technical content.
Subscribe to new posts: Email digest (daily) or RSS feed
Hard system lockups traced to memory exhaustion from loading a 40GB model on a 48GB machine. Three env vars fixed it. Also: real inference validation via smoke check, and why the 70B is off the default toolchain.
An M5 Max MacBook Pro as a local AI inference node: Ollama, 70B and 32B models, unified memory architecture, and a comparison framework for benchmarking local vs. cloud-hosted inference.
A home network hardware migration exposed the fundamental flaw in my RAG system. Here's the architecture Claude and I designed to fix it: temporal state, write safety, and a compounding wiki layer.
An honest postmortem on why an AI assistant repeatedly violated explicit git workflow rules—and why the only solution was to remove the capability to violate them.