Credit risk.

Global experimental lead for YouTube Shopping

Built Google's first causal propensity model to learn "what works for whom" in the context of YouTube Shopping, led to X-XXx in ROI and XX-XX% decrease in program costs by identifying cohorts of creators that can be "nudged" into sustained Shopping success.

Focus areas: experimental design & causal inference.

Causal propensity modelling | stochastic trees | hierarchical causal models

R, Python, and Stan to make data-driven decisions for early stage (Phase I/II) compounds, competitively position assets, and advise on dose schedule using hierarchical modeling and explainable AI / ML.

Led a Python based summer internship on the use of GANs for synthetic patient augmentation, and algorithmic covariate selection in oncology trials.

Bishop Lab for Active Materials

• --

Created an adaptive experimental design framework to maximize the "value" of noisy, nonlinear, hierarchically structured data using probability and information theory.

Exploited conditional entropies, sufficient statistics, and efficient marginalizations to render Bayesian inference and design 10-10,000x faster than a naive implementation to make principled DoE accessible on experimentally relevant timescales.

Shukla Lab for Designer Biomaterials

• --

Engineered an oxygen-permeable, drug-loaded hydrogel designed for burn wound victims to prevent hypoxia / infection of the site. Our therapy provided a much longer term application than conventional therapies, which significantly damage new tissue growth upon removal.

Collaborated with researchers at Harvard University to conjugate metalloporphyrin to computationally quantify oxygen flux, giving doctors a visual indicator of the time-varying oxygen permeability.