Below the

Waterline.

where ideas go before they're ready.

engineers talking shop. implementation notes, architecture decisions, and things we learned the hard way.

React·TypeScript·GSAP·Python·OpenAI·Vite·Tailwind·Node·LangChain·Vercel·Shadows·Oliver·React·TypeScript·GSAP·Python·OpenAI·Vite·Tailwind·Node·LangChain·Vercel·Shadows·Oliver·React·TypeScript·GSAP·Python·OpenAI·Vite·Tailwind·Node·LangChain·Vercel·Shadows·Oliver·

From the log.

What we're building, reading, and thinking about.

August 4, 2026

I turned an ASUS GX10 into our own local model provider

We turned a conference-gift ASUS GX10 into an OpenAI-compatible model provider the whole team can hit with the client code they already use — LiteLLM in front, llama.cpp and vLLM behind, three access tiers, and one genuinely annoying Docker networking wall.

BUILDINGinfrastructurelocal modelsgx10

What we're reading.

Each dot is something worth your time. Hover to see why we think so. Click to read it.

X-axis: practical to theoretical.
Y-axis: engineering to AI/ML.

AI / ML

Practical

Theoretical

Engineering / Infra

Not everything here works.

This is where we test before we recommend. Some become client workflows. Some become Oliver skills. Some get archived. That's the point.

Come join the boat.

We're small. We're building something real. Type 'roles' to see what's open.

bash — keel3-hiring

come join the boat — type 'help' to start