The Stack

Pivots vertical, which is the only way to read a 400-line stack trace without scrolling.

Quiet keys and a scroll wheel that free-spins through a long diff.

The 2026 model, and the only screen here that is not for work. Reading, watching, and staying off the Macs.

Noise cancelling as a focus switch. Hand-off between Mac and phone is the actual feature.

Standing for reviews and calls, sitting for the hard parts.

Lights the desk, not the display. Late sessions cost less.

Local inference for anything I would rather not send over a wire.

For the Postgres work that is faster to click than to type.

Endpoint testing and the collections that end up being the API docs.

The connective tissue under the graphs — retrievers, loaders, model adapters.

Where I go when an agent did something strange and I need the trace, not a theory.

Wraps my tools in MCP so any agent can call them without bespoke glue.

Vector store behind the RAG pipelines. Self-hosted for anything private.

Default for AI service endpoints. Async by default, typed by habit.

Because "it looks right" is not a retrieval metric. Faithfulness and hallucination scoring.

Managed inference when the job needs scale or multimodal input.

Model discovery and fine-tuning. First stop before building anything from scratch.

Exploration lives here; nothing ships from here.

Test, build, deploy. Every repo gets one before it gets a README.

Project management and the docs I will need in six months.

Team coordination, plus the channel my n8n workflows shout into.

Architecture sketching. Most systems here started as a rectangle on this board.

Experiment in letting agents run on a schedule without me in the loop.

Primary cloud — managed ML, compute, and the serverless bits.

DNS, edge caching, and the first line of defence for everything public.

Mesh VPN. Reaches the Mac Mini from anywhere without opening a single port.

Reverse proxy with automatic TLS. Two lines of config per service.

Frontends and this site. Push to main, done.