About Me

I am a Principal Machine Learning Scientist at Prescient Design, Genentech. My research focuses on high-dimensional inference and sampling, with a particular emphasis on developing probabilistic algorithms for active, machine-guided molecular design. I received my Ph.D. in Physics from Stanford University, where I worked on hierarchical Bayesian methods for cosmology. During my Ph.D., I interned at NASA Ames and the Center for Computational Astrophysics at the Flatiron Institute. I hold a B.S. in Mathematics and a B.S. in Physics from Duke University.

Current research focus

  • Prediction-centric Bayesian optimization: Developing “prior-free” Bayesian frameworks with LLMs for decision-making under uncertainty in risk-sensitive settings
  • Robust simulation-based inference: Using physics-driven simulators for likelihood-free inference while accounting for model selection and Sim2Real gap

Recent events

  • July 19, 2025: Invited talk and panel "Targeting the multivariate tails in AI-driven molecular optimization" - The Exploration in AI Today Workshop at ICML 2025 in Vancouver, Canada
  • May 19, 2025: Invited talk "Uncertainty-guided drug discovery" - From Models to Molecules: AI’s Expanding Roles in Therapeutics, hosted by Novoprotein in South San Francisco, CA
  • April 15, 2025: Invited talk "Multivariate tails for active molecular design" - Molecule Maker Lab Institute Symposium 2025 in Urbana, IL
  • November 05, 2024: Poster "Semiparametric conformal prediction of molecular properties" - Molecular Machine Learning (MOML) Conference in Cambridge, MA
  • October 16, 2024: Invited panel "Winning in the age of AI" - 2024 AI Summit in South San Francisco, CA

Research themes

  • Decision-making under uncertainty (AI4Science)
    • Multi‑objective Bayesian optimization for molecular design
    • Productionalizing ML-guided design of antibodies, small molecules, and molecular glues tailored to project-specific desiderata
  • Inference and prediction in high dimensions
    • Semantic uncertainty quantification for LLMs
    • Post-hoc calibration (e.g. conformal prediction, semiparametric methods)
    • Out‑of‑distribution generalization, diagnostics for distribution shifts
  • Sampling in high dimensions
    • Diversity-steered sampling strategies for LLMs
    • Guidance for diffusion-based generative models
    • Sampling algorithms for non‑log‑concave distributions