What Star Formation Taught Me About Solving Business Problems with AI

People sometimes ask how a Ph.D. in astronomy turns into a career in data science. The honest answer is that it's all data. In astronomy, I worked with images of the sky, cleaned large datasets and found meaningful signals, then made decisions about how to analyze what I found. The questions were huge, but the process was surprisingly relevant to the work I do today. The subject matter is different now, but it still starts with complicated data and an open question, then working toward an answer that can stand up to scrutiny.

Making that move from academia into industry wasn't straightforward. A program designed to help academics transition into data science helped me realize the problem wasn't my experience—it was how I talked about it. A company trying to hire you doesn't care that you studied star formation. They care that you know how to analyze terabytes of data. I learned to translate highly specialized academic experience into skills that mattered outside academia: coding, analysis and explaining complicated ideas clearly. That last skill turned out to be especially important: I ran astronomy outreach during graduate school for audiences ranging from elementary school students to families at stargazing events, and explaining science to very different audiences taught me how to make technical ideas understandable without stripping away what matters. I still use that skill with clients today.

Working with clients on AI often starts by resisting the temptation to jump straight to the newest technology. Clients are under pressure to use more AI, particularly generative AI, and I understand why—the promise is compelling: move faster, reduce costs and make teams more productive. But a tool isn't useful simply because it's new.

I'm constantly having conversations with clients that ask: What are we really trying to solve and what's the best way to do it? Keeping the solution as simple as the problem allows is a big part of the job.

The thing that tends to get overlooked in the rush toward AI is data. Clients often think they have more data than they do, or that it's in a more ready state for AI than it really is. Before teams can build useful models or AI-powered products, they usually have to understand what data exists, how reliable it is and what needs to change to make it usable. That can mean starting with one business unit, proving what works and expanding from there. It always comes back to the data.

AI changes constantly, and even as someone who works in this field, I find myself feeling behind. The challenge for me isn't just learning about every new tool—it's separating what's genuinely useful from what's simply getting attention. We're all trying to keep up with the trends, but also discern what's a trend and what's actually really valuable. That makes judgment part of the technical work. Teams still need to develop, test and validate solutions, sometimes with technologies so new that few people understand them deeply. The pace can be exciting. It can also be overwhelming. Both things can be true.

The thing I keep coming back to is how much better the work gets when people with very different skills sit on the same problem together. That collaboration doesn't stop at data science or engineering.

On one project, I worked side by side with a creative colleague to assess images generated by a new tool. The things I thought looked good weren't good. My creative colleague could spot problems I wasn't trained to see, like lighting, texture and positioning. I could adjust the technical side, bring the results back and ask the creative to validate the improvements. Neither of us could have gotten to the right answer alone. That back-and-forth, where someone with a completely different lens challenges your assumptions, is where the best work happens.

Whether the problem is star formation or a client's next AI initiative, the work starts the same way: understand what you're actually trying to solve, make sure the data can support it, then choose the simplest approach that gets you there. The tool matters less than the discipline behind it.