AI and decision support · 2026
How I work with AI
The questions hiring managers ask me about AI, answered with the system I built and run every day. Open any question for the short answer and something to look at, or ask the live demo yourself.
Counts from my own fact files, October 2026.
Ask an AI about my work
It answers only from the public pages of this site, in the third person, and says so when it doesn't know. It runs on a capped budget, so each visitor gets a few questions a day.
Eight questions, with the receipts
01What do you actually use AI for?
Research, drafting, code, design and quality checks, all in one working system. AI handles the format and the logistics. Every fact, story and decision stays mine.
- Research: finding and scoring roles against the same criteria, then confirming each one is live.
- Drafting: résumés assembled from an approved library; letters I review and rewrite.
- Code and design: this site, its access control and this demo, built with AI and directed by me.
- Quality: automated claim checks and a fresh AI reviewer on anything major.
02Show me something you built.
At Innocean I conceived The Brain and built its first prototype: an AI decision-intelligence tool on Microsoft Copilot that I taught to reason with more than 50 industry guides I wrote. Hyundai funded a production module; it was still a prototype when I left. For my own search I built a smaller version of the same pattern, which runs this site and the demo above.
03How do you keep AI from making things up?
Facts are locked; the prose is free. Every claim has to trace to a confirmed row in my fact file, and two checks run before anything ships: an automated claims check and a cold read by a second AI reviewer that hasn't seen the work.
04Who makes the call?
People do. The agent never contacts anyone. I approve every role it scores and send everything myself.
- Facts366 confirmed rows. No row, no claim.
- RulesOne rulebook for tone, length and what never to say.
- AgentFinds, scores and drafts.
- ChecksClaims check, packet audit, cold reviewer.
- MeApprove, edit, send.
05What does it cost, and who watches the spend?
I run it like a budgeted program: a monthly ceiling, unit costs and a ledger. The demo above runs on the smallest capable model with a hard monthly cap, a daily cap per visitor and a stop in the code before the provider’s limit is reached.
- A monthly budget enforced in the code, under a separate limit at the provider.
- A few questions per visitor per day, and a daily cap for everyone.
- Public pages only: no client data, no private case studies.
06Where does AI fall short?
Judgment, taste and context it hasn’t been given. Two examples from this project:
- I approved a design with serif figures; the build shipped them in monospace. I caught it on the first look.
- An AI summary of my employment records suggested listing four separate jobs. The records showed one job under a corporate spin-off, so the summary’s structure didn’t ship.
The lesson I apply: AI drafts fast, and someone accountable checks it against the source.
07What would this look like on my team?
The same pattern I used for The Brain: governed data, written rules and people making the call. I start with one recurring decision the team already makes, lock the sources it should rely on, write down what good looks like, and keep a person approving the output until the checks earn trust.
Be sure you’re right…
Locked sources, written rules and checks before anything ships.
…then go ahead.
Move fast once the guardrails are in place. My dad’s rule, and the reason this works.
08What tools, and how did you learn them?
Claude for the agent, code and design; a second AI model for independent review and imagery; Cloudflare Pages, Access and Workers for the site, the access gate and this demo. Before The Brain, I spent a planned break in 2023 on data-science and AI coursework, on top of earlier training with Johns Hopkins, General Assembly and Google Cloud. I directed these builds; the agents wrote most of the code.
“How you do anything is how you do everything.”
See the work this is built around
The full case studies name clients and show results. Request access with your work email.
Request case-study access Connect on LinkedIn