Making Models Accountable
I technically lead the ML Fairness team at Upstart — a small team focused on studying how our models produce systematically different outcomes across protected groups, and building the technical infrastructure to address that. We've done work on adversarial debiasing of latent representations, Bayesian estimation of disparate impact, and cross-functional policy design for fair lending.
The piece I'm most proud of is a novel post-processing technique for algorithmically debiasing any general ML model, which we applied to Upstart's core underwriting systems. It's the kind of work that demands both mathematical rigor and genuine care about who the models affect — the two things that are most important to me about my work.
Hunting Fraud at Scale
I joined Upstart and moved to San Francisco to help build Upstart's ML Verifications team from the ground up. It was intense, operational, and clarifying: this is where I started to really understand what it means for a model to fail, and what's at stake when it does. My work ranged from designing fraud detection models, to building causal inference frameworks to optimize how we use those models, to ultimately creating monitoring systems that make sure the models are doing what we want them to.
Applying Models in the Real World
After graduating, I moved to Boston to join Splice Machine as a data scientist. I spent most of my time designing Bayesian time series models to forecast patient health trajectories — work that sat right at the boundary between the academic modeling I'd done and the messy realities of production data. It was the transition I needed: learning what it actually takes to build something that has to work every day.
What is Data Science
Between degrees, I spent a summer at MilliporeSigma (Merck KGaA) in St. Louis. It was my first time applying data science inside a large organization — building statistical models to characterize customer health. Small scope, but it was enough to know that I wanted to keep building things outside of academia.
Finding the Right Questions
I came to Washington University in St. Louis to study systems engineering, and along the way I picked up my Masters in analytics and statistics, and a minor in computer science. But the work that shaped me most was research in mathematical epidemiology: building models for HIV/AIDS to study how prevention methods like PrEP affect epidemic stability. It was the first time I experienced what it felt like to build a model that mattered and showed me how I could help answer important questions with data.
Read the paper →