01
AI infrastructure
Local inference & serving
Running models on my own hardware — llama.cpp and vLLM on a GB10 desk-side node — and building the routing, health checks, and evaluation loops around them.
About · San Francisco
Practical systems, built to be operated.
I'm Nathan — a systems engineer focused on the layer between hardware, AI models, and the people operating them. Most of what I build runs in my own lab: local inference, automation, and observability that has to actually work, because I use it daily.
How I think about the work
Infrastructure is most interesting when people can actually use it.
My background is in systems engineering and infrastructure. I care less about demos and more about the unglamorous parts that make software dependable — deployment, telemetry, failure modes, and the operator experience around all of it.
The lab on this site is a small production environment, not a mockup. Local models get served, automated agents run on real schedules, and monitoring watches nodes that fail in ordinary ways. If a system can't survive contact with my own daily use, I keep working on it.
Focus
What I work on, and how.
02
Automation
Agentic operations
Automation and agent frameworks on real schedules with real side effects: pipelines that discover, evaluate, and act — with checks and validation at each step.
03
Observability
Telemetry & operator tooling
Local-first observability across the lab: health monitoring, dashboards, and the operator-facing surfaces that make the rest legible.
04
Systems engineering
Prototype, then harden
Prototype quickly, pressure-test against real usage, and keep the interfaces clean enough to rebuild. Boring reliability is a feature.
Tooling
Systems I work with.
Inference & models
Infrastructure
Data & state
Observability & ops
Elsewhere on this site
The projects and the lab tell the rest of the story.
Everything referenced here is documented on this site — built, deployed, and operated by me.