I am a data scientist building and testing foundational AI at Spotify.
At Spotify, I help bridge research and product on a foundational LLM trained on user listening — mostly by owning evaluation, from offline metrics that reflect a model's behavior across tasks to the A/B tests that put it in front of listeners. I spend my time on semantic IDs, synthetic data, LLM-as-judge, online–offline correlation, and safety.
The pull back to public-interest work is strong. At the City of Boston I partnered with a range of departments to tackle challenges including COVID response, shelter reform, and rideshare pickup zones. At Oceana, I modeled and mapped overfishing and illegal fishing. The data science toolkit, pointed at the right problems, is a game-changer in sectors that are late adopters.
I still miss jets. My PhD at Columbia was on jet physics — years optimizing jet algorithms, culminating in a supersymmetry search in collisions featuring an unusually large number of them. I also loved the hardware side, testing and validating ATLAS's Pixel detector. Now, from the outside, I'm fascinated by how thoroughly a decade of AI has reshaped how high-energy physics actually gets done.
Alongside my day job I've taught graduate-level adjunct courses at Northeastern, BU Questrom, and Harvard Extension School, specializing in data visualization and storytelling. Always open to conversations about new courses or guest lectures.
European Commission on the Environment, Marine Biogeographic Seminar
PhD Thesis — Columbia / ATLAS, CERN
arXiv:1602.06194
arXiv:1510.03823