Blog

Building a solver for the NYT Connections puzzle turns out to be a hands-on lesson in the philosophy behind similarity search, embeddings, and NLP — meaning as geometry, and why polysemy breaks it.

Why I write agent skills: they capture the things I type constantly, scope the context an AI actually needs, pin down process for a whole team, and reach for live context through MCP servers.

A broad, practical walkthrough of training a chess model to imitate one player's style, and why it is a useful way to understand fine-tuning.

Meta, Scale AI, workplace monitoring headlines, the two clocks problem (state versus event), and why "replicating human expertise" bumps into what our systems never stored.

Why standardizing Tableau workbook deployments matters, how TabReview automates workbook analysis, and where embedded BI governance is heading.

How running JupyterLab in the browser with JupyterLite can shift costs from your JupyterHub cluster to the client — with napkin math, limits, and a practical hybrid deployment approach.

What watching an AI debugger reveals about where debugging methodology actually comes from — and what it means that the best teacher learned from the room.

Prediction markets have reopened arbitrage to anyone with infrastructure and a wallet. What that means, how it works, and where the legal lines are.

At what point does a Mac Mini running Claude replace a $50/hr employee? An honest look at the numbers behind AI agent economics.