I'm a data scientist interested in what happens when algorithms leave the whiteboard and enter the world: how they are built, what assumptions they encode, how they behave once deployed, and how their performance and consequences can be evaluated scientifically. My work draws on artificial intelligence and machine learning, statistics, mathematical modeling, optimization, and the building and analysis of complex software systems. In practice, I move back and forth among code, data, and equations. I read systems closely, build models from large and often messy datasets, and use each source of evidence to understand the others.

The question behind much of my work is deceptively simple: What does the system actually do? Sometimes answering it means reconstructing an algorithm from source code. Sometimes it means building a mathematical model of its behavior, testing that model against data, or measuring the effects it produces in practice. Often it requires all of these at once.

I am especially drawn to problems in which algorithms interact with markets, institutions, and human decisions, where a model is not merely a mathematical object but a working part of the world.

Much of my applied research happens at Analysis Group, where I lead technical teams working on difficult quantitative questions in litigation and applied economics. The work ranges from the forensic analysis of complex software systems to large-scale statistical and mathematical modeling, as well as the design of NLP and machine-learning methods.

It has taken me into algorithmic collusion and antitrust, evolutionary algorithms that design neural-network architectures, and the forensic task of reconstructing what software did from the code and data it left behind.

Some of that work eventually has to travel from code, data, and equations into the courtroom. That translation is part of expert witness work. It requires more than presenting a conclusion. It requires making the path to that conclusion visible, including the assumptions, the tests, and the evidence on which it rests.

To me, the value of that work lies in bringing the habits of scientific inquiry to questions that courts must decide. The contribution of science in that setting is to make technical claims examinable, to separate intuition from evidence, and to show clearly what the facts support.

Lately, I have been thinking about what agentic AI systems can actually do inside serious quantitative work, not merely what they can automate around it. One thread I keep pulling on is the use of large language models as measurement tools within statistical and mathematical modeling: extracting structure from unstructured information, validating what they produce, and understanding how their outputs affect the models built from them. To me, that feels like a genuine shift in how applied quantitative work will be done.

I'm also an adjunct professor at Boston College, where I teach graduate courses in machine learning, econometrics, programming, and data analysis. The mathematics underneath these subjects is too often taught as something dry and abstract: equations on a board, detached from what they actually do to data. For the record, I love a good board full of equations. But I also want students to see the machinery move.

This lab grew out of that impulse. It is a collection of tools you can manipulate and watch respond: a coefficient you can move, a dataset you can resample, or a loss surface that changes under your hand. The aim is for the ideas to stop being merely symbolic and become things you can see, test, and reason about directly.