Curriculum Vitae

Ph.D. in Physics

Princeton University, Princeton, New Jersey, United States

• Advisor: Jo Dunkley
• Research interests: Statistical inference, Monte Carlo methods, machine learning, simulation-based inference, cosmic microwave background, component separation, and Galactic foregrounds.

M.A. in Physics

Princeton University, Princeton, New Jersey, United States

B.S. in Applied Mathematics & Statistics, Physics, and Mathematics

Johns Hopkins University, Baltimore, Maryland, United States

• GPA: 3.99 / 4.00 (General Honors and Departmental Honors in all majors)
• Minor: Computer Science

• Donald E. Kerr Memorial Award, JHU top 2 graduating seniors majoring in physics
• Applied Math and Statistics Achievement Award, JHU top 6 graduating seniors majoring in applied math and statistics

• Sigma Pi Sigma Physics Honors Society
• Provost's Undergraduate Research Award, JHU, $6,000 in funded research support

• Bloomberg Distinguished Professor Research Fellowship ×2, JHU, $10,000 in funded research support

• Dean's List

Graduate Researcher | Simons Observatory

Princeton, NJ | Mentors: Jo Dunkley, Susanna Azzoni

• Led a first-author study of model misspecification in cosmological inference using cross-spectrum component separation and realistic simulation ensembles.
• Developed regression and template-fitting estimators for multi-frequency polarization data, including weighted cross-template fits, harmonic filtering, and Monte Carlo uncertainty propagation.
• Designed Bayesian/MCMC and simulation analyses to quantify estimator bias, uncertainty, covariance, chance correlations, and robustness under foreground model misspecification.
• Built SO×Planck cross-correlation and calibration pipelines to validate instrument performance and foreground measurements against independent data.

Undergraduate Researcher | Cosmology Large Angular Scale Surveyor

Baltimore, MD | Mentors: Charles L. Bennett, Tobias A. Marriage, Ivan L. Padilla

• Implemented and optimized a needlet-domain minimum-variance estimator for multi-frequency signal decomposition.
• Developed foreground masking and Fourier-space point-source detection methods and benchmarked estimator performance across frequency bands.
• Built Python diagnostics for temporal anomaly detection in observational time-series data using rolling-dispersion statistics.

Undergraduate Researcher | Zakamska Astrophysics Group

Baltimore, MD | Mentors: Nadia L. Zakamska, Hsiang-Chih Hwang

• First-authored a peer-reviewed study modeling a compact binary system using spectral decomposition and time-series fitting.
• Developed automated Python pipelines for photometric and spectroscopic feature extraction in the time and frequency domains.

Benchmarking Scientific Reasoning in AI Agents | Terminal-Bench-Science

Terminal-Bench-Science

• Design executable scientific tasks that test whether AI agents construct models that generalize beyond visible evaluation constraints.
• Authored a CMB inference benchmark targeting model misspecification and restrictive assumptions that can remain statistically adequate on observed data while failing under hidden evaluation conditions.
• Review astronomy and physics tasks for scientific validity, difficulty, verifier robustness, reproducibility, and potential shortcut solutions.

Representation Learning and Simulation-Based Inference for Cosmology

• Developing VICReg-based self-supervised representations of simulated CMB observations for downstream inference of the tensor-to-scalar ratio r.
• Investigating learned low-dimensional summaries that retain cosmological parameter information while remaining robust to foreground and observational variation.
• Exploring simulation-based inference for likelihood-free posterior estimation using learned representations.

Reinforcement Learning for Sequential Auto-Bidding

AuctionNet · COS 435 group project, Princeton University

• Developed PPO agents for sequential decision making in a replay-based auction environment subject to budget and cost-per-action constraints.
• Extended the policy with temporal observation stacking and recurrent representations to capture nonstationary market dynamics and long-horizon dependencies.
• Evaluated reward shaping, market-state features, and alternative policy architectures through controlled experiments and ablations.

Transformer Inference Systems Benchmarking

• Designed controlled experiments to study the compute–memory tradeoff of KV caching in decoder-only transformers across prompt and generation lengths.
• Benchmarked decoding throughput, end-to-end latency, cache footprint, and GPU memory behavior under cached and non-cached inference.
• Characterized cache-memory scaling, prefix reuse, CUDA allocator behavior, and performance degradation under CPU–GPU memory offloading.

Simons Observatory: Assessing and Modeling Foreground Complexity in CMB Analyses

Oral presentation · 2026 American Physical Society Global Physics Summit, Denver, CO

The impact of dust complexity on the recovery of r

Invited talk · Pan-Experiment Galactic Science Group, Los Angeles, CA

From the Impact of Dust Complexity to Beyond

Invited talk · 2025 Simons Observatory Collaboration Meeting, Philadelphia, PA

Understanding the impact of dust complexity on the recovery of the tensor-to-scalar ratio

Oral presentation · 246th Meeting of the American Astronomical Society, Anchorage, AK

Understanding the impact of dust complexity on bias in r

Invited talk · 2024 Simons Observatory Collaboration Meeting, Chicago, IL

CMB analysis using the global-NILC method

Poster · 2023 American Physical Society March Meeting, Las Vegas, NV

CSS1603+19: a low-mass polar at the cataclysmic variable period minimum

Poster · 241st Meeting of the American Astronomical Society, Seattle, WA

Teaching Assistant, EN.553.430/630 Intro to Statistics

Johns Hopkins University · Undergraduate/Graduate, 82 students

• Graduate-level introductory statistics course covering stochastic convergence, point estimation, hypothesis testing, and interval estimation.

Teaching Assistant, EN.553.633 Monte Carlo Method

Johns Hopkins University · Graduate, 60 students

• Graduate-level course covering pseudorandom number generators and classical Monte Carlo algorithms including Metropolis–Hastings.

Teaching Assistant, AS.171.103 General Physics I

Johns Hopkins University · Undergraduate, 23 students, 2 semesters

• Undergraduate-level introductory physics on classical mechanics.