Operator first.

I am Mark J. Salisbury. I live in northeastern Connecticut, and I build, test, and teach practical AI.

I started behind a Geek Squad counter. Eleven years of diagnosing, fixing, and explaining technology to people who did not care how it worked and should not have had to. That job taught me the thing I still use every day. If the person in front of you cannot run it after you leave, you did not fix anything.

After that came a decade in operations. Digital billboard inventory, vendor management, construction, a custom enclosure manufacturer. Different industries, same problem every time. The real process lived in one person’s head and a spreadsheet nobody was allowed to touch. I got good at finding that spreadsheet, understanding why it existed, and building something better around it.

When the AI tools got good, I did what I always do with a new tool. I put it to work on real jobs, broke it on purpose, and wrote down what held up. That turned into StarForge Dynamics, the training and advisory company I founded in 2024. It turned into TeamLoop, an open-source toolkit for running several AI models as one small accountable team. And it turned into teaching, which is still my favorite part of the work.

Then there is the part I do purely for fun, which is design. Web pages, interfaces, the look and feel of a thing. Every project I build gets its own face because I cannot leave a default alone. I do not make websites for a living. I make them because I love it, and I am good at it. The design page has the evidence, and tells you how to ask for one.

Today my day job is AI Outcomes Manager at Glean. This site is separate from that work and says nothing about it. What you see here was built on my own time, with my own accounts, because I cannot stop building things.

How I work

• Start with the job. The workflow comes first. The model gets picked later, and sometimes the answer is no model at all.
• Near-correct is the dangerous kind of wrong. A result that looks right gets less scrutiny than one that looks broken. So I check the good-looking ones hardest.
• Keep the failures. A record with only the wins in it is marketing. I publish the ugly runs, including the ones where my own setup was the problem.
• Put the hard guarantee outside the model. Capability is not authority. If something must be true, a person or a plain piece of code is responsible for it.
• Teach it or it does not count. A tool only I can run is a liability I created.

The short version

For introductions, programs, and bios. Use it as written.

Mark J. Salisbury is an AI operator, builder, and teacher based in Connecticut. He is the founder of StarForge Dynamics, the creator of the open-source TeamLoop toolkit, and an AI Outcomes Manager at Glean. He builds real tools with AI, publishes how he tests them, and teaches people the parts that hold up.