About
I am a data scientist in the Data and AI Group at Bank of America in New York, where I lead applied research for an agentic AI platform. Before joining the bank, I spent five years at Lawrence Berkeley National Lab developing machine learning methods for particle physics, and I hold a PhD in Physics from UC Berkeley (2025).
Current Work
At Bank of America, I lead the applied research vertical of a team building an agentic AI platform. My work focuses on two questions: how to build AI systems that can plan, act, notice when something has gone wrong, and recover; and how to evaluate those systems rigorously, by what their outputs actually do rather than by how plausible they look. Alongside the research, I build and ship the production retrieval and evaluation systems the platform runs on.
Research
My research has always been about inference under uncertainty: recovering what is really there from data that has been distorted, smeared, or corrupted by the process of measuring it. In particle physics, that problem is called unfolding. Collider detectors blur every measurement, and turning those distorted observations back into statements about the underlying physics is one of the field’s central statistical challenges.
At Berkeley (2020–2025), working with Professor Benjamin Nachman at Lawrence Berkeley National Lab, I designed machine learning methods to push the limits of the information that can be extracted from particle physics data. My dissertation, “Machine Learning Methods for Cross Section Measurements,” develops a framework for applying modern generative and adversarial models to these measurements. Highlights of my researchinclude:
- Moment Unfolding
- A GAN–inspired method that unfolds the moments of a distribution directly, without histogram binning, enabling more precise comparisons with theory (Physical Review D, 2024).
- Reweighting Adversarial Networks
- Generalizes Moment Unfolding to full phase-space unfolding, using a particle–level reweighting function steered by a Wasserstein critic. RANs remain stable even when particle– and detector–level distributions barely overlap, and outperform state-of-the-art methods at lower computational cost (Physical Review X, in press).
- Neural Posterior Unfolding
- Uses normalising flows and neural posterior estimation to perform unfolding, with implicit regularisation from the network and fast, amortised inference (Journal of Instrumentation, 2026).
- Unbinned Inference with Correlated Events
- Shows that unbinned inference on unfolded data breaks a standard assumption, that events are statistically independent, and that ignoring the resulting correlations can significantly underestimate uncertainties (European Physical Journal C, 2025).
- Unfolding in the Presence of Nuisance Parameters
- Extends the OmniFold algorithm so that machine learning–based unfolding can profile the nuisance parameters that encode an imperfectly known detector model (Annals of Applied Statistics, in press).
- SymmetryGAN
- A deep learning method that automatically discovers the symmetries of a dataset, with applications from particle physics to broader data science (Physical Review D, 2022).
The same thread runs through my work on AI systems today. A measurement is only as trustworthy as your understanding of how it was distorted, and the uncertainty it carries is part of the result. Evaluating an agent by what its outputs actually do asks the same thing of it that unfolding asks of a collider detector.
My physics research has appeared in venues including NeurIPS (2021, 2022, 2024), Physical Review X, Physical Review D, the European Physical Journal C, the Journal of Instrumentation, and the Annals of Applied Statistics, and I have given invited talks at CERN, the Korea Institute for Advanced Study, NeurIPS, and the American Physical Society. More details about my research and talks can be found on the publications and talks pages.
Education
I earned my PhD in Physics at UC Berkeley in 2025, advised by Professor Benjamin Nachman. My dissertation, “Machine Learning Methods for Cross Section Measurements,” develops a framework for applying modern generative and adversarial models to these particle physics measurements.
For my undergraduate studies, I attended Yale, where I double majored in Mathematics and Physics (BS with Distinction in both) and completed a simultaneous MS (Mathematics) in 2020. At Yale, I was awarded the Howard L. Schultz Prize for the most outstanding graduating senior in physics. As part of my undergraduate research, I published papers in pure mathematics and theoretical physics.
Industry Experience
Before Bank of America, I worked across several domains that share the same underlying problem of extracting reliable signal from noisy data:
- Emissary AI (2025)
- Fine-tuned language and vision-language models for client applications, including a GRPO–based reinforcement learning pipeline for code completion and image–based IP violation detection, and multi-GPU distributed training.
- Bridgewater Associates (2023)
- Built Bayesian hierarchical models to forecast liquidity and transaction costs for trade execution, using partial pooling to make reliable predictions from sparse data.
- Microsoft Research (2022)
- Worked with Jaron Lanier on nonlocal field theory and matrix models, developing operator-theoretic and stochastic methods that bridge discrete and continuous optimisation.
Service
I review for Archives of Computational Methods in Engineering, Nature Scientific Reports, the Journal of High Energy Physics, and NeurIPS, and served on the UC Berkeley Physics Faculty Search Committee (2021–2024). I am a member of the Sigma Xi Scientific Research Honor Society and Sigma Pi Sigma, the physics honor society.
Outside Work
I’m drawn to thinking deeply about the logical structure of the world around us, and that curiosity doesn’t stop at physics or machine learning. I’m happy to talk with anyone about anything from linguistics to molecular biology to computer architecture. Away from the desk, I play badminton and the piano, and enjoy running.
I’m always open to research collaborations. If you have an idea you’d like to work on together, please get in touch.