I did not come to AI from research. I came up through credit union operations, running core conversions overnight, answering examiner findings, and keeping the technology at a nearly $8 billion institution reliable enough that nobody had to think about it.
My first job out of college was building web applications for hotels and golf resorts. I moved into credit unions in 2007 with a nine-person IT team at a $450 million shop in Maryland, and spent the next two decades working through the same problems at bigger and bigger scale: data centers, then core platforms, then digital banking, then data and security, and now AI.
That history shapes what I pay attention to. The hard questions in this industry are rarely about what a model can do. They are about whether you can explain a decision to a member, show an examiner the control that governed it, and still move fast enough for any of it to matter.
What I actually do
At Opmodi I lead the AI Center of Excellence, funded by credit unions for credit unions. Most institutions cannot justify a full-time AI function on their own. Together they can fund a good one, and my job is to make that worth paying for.
Most of the work is finding out where an institution actually stands, which is rarely where its vendors say it stands. After that comes the governance that makes deployment defensible: model inventories, risk tiering, approval gates, oversight. The part people skip is staying close enough to delivery that none of it turns into a document nobody opens.
I have done that part from the inside. At Mid-Hudson Valley Federal Credit Union I wrote the AI policy, then built the system controls that enforced it and the tooling that kept member data out of places it should not go while engineering teams used AI to move faster. A policy nobody can enforce does not do much on its own.
The unglamorous part
Most of the durable value here is boring. Data lineage, access control, token costs, model routing, and finding out which vendor has quietly been sending member data through a model nobody approved. A fair amount of it is deciding not to build something.
I have led award-winning AI work, and the award-winning part was never the model. It was that the institution was ready for it. The cloud migration was done, the data was in decent shape, and compliance had been in the room since week two instead of week twenty.
The Symitar conversion I ran at Hudson Valley in 2015 is not a good story. It saved over half a million dollars a year and nothing broke. I count that as the best kind of project.
Why credit unions
Credit unions are cooperatives, which means the members own them. That changes how I think about risk. An AI failure at a bank is a bad quarter. Here it is a broken promise to people who joined because they trusted the place.
It also means competing against banks with far bigger technology budgets. Shared infrastructure is how credit unions close that gap without every one of them learning the same expensive lessons separately. That is why I took this job.
Outside work I chair the board of Family Services, a community services organization in the Hudson Valley, and serve on the board of Arts Mid-Hudson. Board work is a useful reminder that governance is something people practice, not something you write down once.