Alex Lentz

I'm a data scientist focused on building AI systems that work in practice — not the slick demo, the version that holds up when real users push on it. I care less about how clever the model is and more about whether it actually does the job it was built for.

These days I lead the data science team at an EdTech company. Before that, I spent years consulting — automated predictive modeling, customer segmentation, forecasting, visual analytics, the occasional training class — across media, government and non-profits, a political polling firm, and a world-renowned hedge fund. Sounds eclectic until you notice every one of those clients was trying to do the same thing: make a decision under uncertainty. The data changed; the problem didn't.

I didn't plan any of this. I started in politics and economics, worked on campaigns and in operations, and kept running into the same realization: the people with the best answers were the ones closest to the data. So I went all in — first at NC State's Institute for Advanced Analytics, then UPenn's MCIT program while running a team full time. I'm still slowly chipping away at an MSE in Artificial Intelligence at UPenn — one course at a time, on the side.

Lately a lot of my attention is on the AI side of the field — agents, evals, retrieval, and figuring out where LLMs actually help versus where they just feel impressive in a demo. I'm less interested in the next benchmark number and more interested in what a thoughtful human-plus-model workflow looks like for a student, a parent, or an advisor.

The further I get into this work, the more I think the hard part is judgment — knowing which problems are worth pointing a model at, what "good" looks like, and when to throw the whole thing out and try something simpler. The math is the easy part now. Everything around it is the job.

Outside of work: family, "old man" baseball, beer league hockey, the NC outdoors, an unreasonable obsession with custom keyboards, and cheering on the Canes or my college teams.