From the work
Writing
Essays written from the work: AI and agents, networking, and the legacy-modernization playbook, with the code and the tests where the claims are.
- Four traps in hospital price data Jul 2026 The schema parsers are free; nobody reconciles what's inside. What building that layer over one metro required: invented numbers, mislabeled rows, corrections that manufacture the bug, and comparisons no correctness check can see.
- the $25 AI lab Apr 2026 Fifteen AI models, a $25/year VPS, a URL-shortener challenge. Eight passed, and most of the failures were the providers, not the models. Then a code audit found that "passing tests" doesn't mean the same thing for every model.
- Network Tools Trilogy Apr 2026 netgrep, termshark, wiregraph. Three tools for reading network traffic, built in Rust and Go because the standard ones give up first. All FOSS.
- six problems, six languages, one AI Mar 2026 Six coding exercises in six languages, 968 tests. The languages and the architecture were mine; the AI wrote the code. What that actually takes.
- Fracture: a 2012 Mozilla demo into a live multiplayer RPG Mar 2026 Mozilla's abandoned 2012 BrowserQuest, rebuilt into a live multiplayer RPG: 3,161 tests, distributed tracing, AI-powered NPCs. The legacy-modernization playbook, applied to a game.
- building a digital doorman Mar 2026 An AI agent on a $7/month VPS that answered questions about my work by reading the actual code over IRC. The architecture mattered more than the model. Since decommissioned; the write-up holds.
- Hermes Agent: Honest Review Mar 2026 Paid $10 to test NousResearch's Hermes Agent. The multi-platform gateway is real engineering; the model is a Llama fine-tune. Who should look, and who can skip.