Where the Data Can't Leave: Self-Hosted AI for Regulated Industries

Live sessions

Learn directly from Thomas Underhill in a real-time, interactive format.

Your Own Reference Architecture

You do not leave with notes. You leave with a costed reference architecture for your own environment: the objection-to-control mapping, the model and hardware tier, the deployment and audit practices, and a total cost of ownership case. If you took the hands-on track, it has evidence attached in an appendix rather than intentions.

Run It On Your Own Machine

In week one you install a capable model on your laptop, disconnect from the network, and keep working. No cloud spend, no GPU purchase, no vendor involved. Most people in regulated industries have never seen this work, and twenty minutes of watching it run changes what the rest of the course means.

1:1 Session with Thomas

One 30 minute session on your architecture, bookable from week three once you have a draft worth reviewing. Bring the part you would rather not raise in front of the group: your regulator, your data, the objection nobody has answered. I have built this stack and will tell you plainly whether your design holds.

Group Office Hours, Off the Record

Two optional sessions, deliberately not recorded, so the room can be honest about what is actually blocked and why. This is where a hospital CTO says the real problem was the auditor's third question, and a bank CISO says the same thing happened to them. That conversation does not happen on a recording.

Post-course access

Go back to course content and recordings whenever you need to.

Certificate of completion

Share your new skills with your employer or on LinkedIn.

Peers You Can Actually Ask

A closed channel with the people who did this work alongside you, still there after the course ends. Healthcare, finance, government and legal, the same problem under different regime names. Nowhere else can you ask how someone answered their auditor without disclosing a gap in your own program.

Two Ways to Do the Hands-On Work

Every week has an optional exercise and you choose how to take it. Build track: point a local coding assistant at the week's problem and construct it. Review track: I hand you working code with realistic flaws planted in it, and you find them. Switch week to week.

It Does Not End at Session Eight

Two things arrive after session eight. In December, a lesson on how to tell whether what you built actually works, with the exercise (model evals) that measures it. In January, a sixty minute alumni session, not recorded, where you come back and say what happened, including if the answer is nothing. Both free, and both are why people stay in touch.