The Story.

First Steps

I grew up in Schaumburg, northwest of Chicago, and started programming in high school. My first course hooked me as a sophomore: we started in Alice, then coded Space Invaders and Frogger clones in Visual Basic. AP Computer Science A as a junior taught me object-oriented fundamentals in Java, and a senior-year mobile development course ended with an iOS game written in Swift in Xcode. Each course raised the stakes a little, and by the end I knew I wanted a career in computer science somehow.

Undergrad at UIUC

I went to Illinois as a pre-engineering student aiming at computer science and started taking CS courses immediately. Plans changed, and I switched to actuarial science. It wasn’t the degree I’d originally planned on, but it gave me a strong foundation in applied math and statistics. A statistical learning course in R was the first time machine learning felt like something I could see myself doing for a living. Three years and three actuarial exams later, I finished my bachelor’s degree and landed my first full-time role.

Actuarial Analyst at Zurich

I joined Zurich North America on the Workers’ Compensation pricing team. The day-to-day was pricing datasets, statistical analyses, loss development studies, and a reporting scorecard built in SQL, R, and SAS. That first year gave me a solid foundation, and because my projects kept putting me next to the team’s data scientists, my interest in an alternate career path grew steadily.

Transition to Data Science

I applied for an internal move to the pricing quant team and joined as an associate data scientist. I wanted to focus on computing, so I stopped taking actuarial exams and started considering graduate programs. The work carried over naturally: as an analyst I had prepared the lookup tables and input datasets supporting the Workers’ Compensation pricing model rebuild, and as an associate data scientist I used them to help train the models themselves. One of my first projects was porting a core modeling pipeline from Pandas to PySpark on Databricks and cutting its runtime from five hours to an hour and a half.

Production Pricing Models

I was promoted to data scientist in October 2023. I rebuilt the Medical-Only injury-type submodel of the Workers’ Compensation pricing model, a Tweedie GLM, and built a parallel Databricks batch scoring pipeline alongside it. The production scoring itself lives in Zurich’s internal pricing platforms, built by our platform engineers from the model coefficients and feature specifications; my pipeline matches it nearly one to one, which lets data scientists batch score policies for model price review studies and is intended to shorten the next rebuild.

I also owned submodels in the US General Liability refresh, re-engineered Zurich Canada’s General Liability model for the company’s re-entry into that market, and presented model-performance reviews across Workers’ Comp, General Liability, and Auto, including models I had not built myself.

From iCAN to the MCS

I went back to Illinois while working full time. First iCAN, the computing accelerator built for people whose bachelor’s degree is in something other than computer science, then the Master of Computer Science itself through the Chicago-based MCS program, with a small in-person component downtown. Distributed systems, applied machine learning, AI agents, and a summer of human-computer interaction research. The distributed systems course produced a stream processor I wrote from scratch in Go, which remains the hardest and most satisfying thing I have built.

AI Document Understanding

I moved onto Zurich’s AI Document Understanding team, where I lead development of an underwriting sustainability agent: it reads a company’s public disclosures against our global sustainability guidelines and rates each provision red, yellow, or green, with every verdict cited back to its source.

I built the full-stack application and initially deployed it with an Azure Function App backend and a simple HTML frontend. I also built a parallel Streamlit version for fast local iteration. Integration with internal pricing and quoting systems is still being considered. There is also a second assistant that brings structured account data into a middle-market underwriting guidelines workflow.