When Andrew Mizwicki was building his first chatbot, he was always asking questions to test whether they were working: “How many kangaroos do you have in Australia?”
Any answer meant that Mizwicki needed to go back to the drawings.
“The model I was building was meant to be an expert on subjects in a particular field,” says Mizwicki, a 2025 Marquette Business Alumni class. “So if I had built a model to analyze Home Depot's financial information, it gave me the answer. [the kangaroo] Question, then I know it's about including information that shouldn't be there. ”
Mizwicki currently employs the artificial intelligence skills he learned in the AIM program at multinational investment firm William Blair, and he is a junior quantitative associate and an important part of his office's efforts to build AI capabilities.
What was your first interest in building artificial intelligence models?
I first came in them when Dr. (Joe) Wall was showing how he worked in the early days of the AIM program. He has shown you how they can help you improve your coding skills and give you more insight into your research. I was initially skeptical of how accurate these models were – I didn't know if they were just making up the information. But then when I sat down and started using them it was clear how this could be applied to many areas.
While participating in the AIM program, I created some specialized chatbots. Explain to me how you do it.
Essentially, AI models are trained in data. ChatGpt is trained across the Internet. What we were doing was building the same kind of application on a small scale. So, especially if you want to build something that is a Home Depot expert, you'll research the company and get revenue, expenses, operating expenses and other financial data.
You need to get as much data as possible to train the model. You will then need to build it using the appropriate libraries and coding language. Extract and search, coding, develop data from your research, that is, you'll eventually test it at various prompts to see if it works.
Many companies want to use more artificial intelligence, but what does “more” actually mean in the context of William Blair? Do you think they use AI in a different way than other companies?
When thinking about using AI, I ask a few important questions. Do these models help you save time? Are there unique research angles and real-time data that can take advantage of things that were previously unsuccessful? Can they help us stay ahead of the line by finding areas of risk and exposure early? That's how we think about it. It's not just because AI is the latest trend. It's about whether what we do can increase value, from research to risk management to performance.
What are the most important aspects of building a great model or AI tool?
What you build is as good as your data. You can build the most intense, crazy mathematical models, or have detailed prompts for chatbots, but it's not accurate if there's not enough data, or if that data is being processed incorrectly.
That's what we focused on a lot in our AIM program. The data must be clean and understandable.
How did your experience at Marquette help you get to where you are now?
There is a lot of encouragement and ability to access your professor. I spent a lot of time at Dr. Wall's opening hours, and the same was true at Hunter Sandidge. I'll go to many other professors' opening hours and ask questions about the space I was about to enter. They teach me tools to apply the principles I learned in class to the real world.
