Case studies & war stories
A whitepaper turned into a slide deck. A learning plan for microcontrollers.
My day job: I lead the Public Sector business unit at Straive, and I'm CEO of Double Line — the company we built and sold, now a subsidiary there.
This site is about everything else. I take things apart to understand them — a retirement model, experiments with LLM internals, the Godot binary behind a game I was curious about. Most of it starts as a question I can't answer and ends as something that runs.
Lately the questions are all about AI, and overall my observation is: the people who understand how these models work have never run a business, and the people running businesses use these models daily without knowing how they work. I'd like to be a third thing. Writing down the attempt is what the Lab is for.
The AI Lab
Case studies with the failures left in, the prompts themselves, and the paths I actually followed.
What persists between conversations and what doesn't, how tool use works, what RAG and embeddings are for, token economics, evaluation — the nuts and bolts that are too often glossed over. I'm not trying to fix the ML engineer - business gap, just wanting to not fall prey to it.
Research & analysis
The LLM psychology experiments. When Claude detects silent disagreement.
Prompts & assets
The learning‑plan generator, as a prompt and as a Claude skill. Yours.
Currently, honestly
Passed the Claude Certified Architect exam in July 2026. Now: hallucination detection and output validation.
Things I've built
Handled
liveAudit‑grade time tracking for R&D tax credit documentation — built so CPA review is straightforward instead of a scramble at filing time.
Well Spent
liveRetirement tools model running out of money. Almost none model the more common ending: dying with a decade of unspent work in the account. This one leads with that number, if for no other reason than to add some weight to the underspending discussion.
The AI Alternatives Project
ongoingWhich SaaS categories AI can genuinely replace, and which ones only demo well. With a companion bounty board where developers submit their attempts and their Claude Code folders.
Writing worth your time
Who should hold a district's data, and why vendors resist interoperating
The argument the AI Alternatives Project came out of
States are investing; the Pre‑K to K‑12 join is the part nobody plans
Conversations
Helping states make smarter use of their data · Sep 2025
Developer to CEO, and the mistakes in between · Dec 2023
Reframing failure · Nov 2023
Where I've been
Double Line since 2012 — data engineer first, then project manager, director, VP of
technology, COO, then CEO. Deloitte Consulting before that.
Professional LAMP for Wrox in 2005, a Google Summer of Code fellowship the same
year, systems analysis and psychology at Miami University.
The full CV →
Off the clock
I earned my pilot certificate in November 2017 — I expected the training to be difficult
and it was harder than I thought. Travel, mapped. And a reading list currently dominated by
neural networks, though the pile next to the bed is more honest than that sounds.