Wrote end-to-end, production-ready statistical software and supporting dashboard hosted on an AWS EC2 instance to identify drivers of topline metrics and test for mix-shifts, largely automating the creation of callouts for Weekly Business Reports. Separately, used Amazon Bedrock for prompt-engineering for LLM-generated SQL to improve Claude-powered Text2SQL; employed mixed effect modeling to identify points for improvement.

Reduced churn from Amazon Currency Converter using parametric modeling (Cox proportional hazards (CPH) with time-varying covariates), accelerated failure time (AFT), multi-state models, as well as packages PySurvival and Scikit-Survival; provided richer business-actionable information compared to classification approaches to churn, including (1) probabilities of progression to the early warning signs of churn, (2) seller disbursement dollar amounts at risk, and (3) causal inference on variables driving churn, calling out the troubling features of each seller.

At DevIntel, built metric-monitoring system using Python targeting changepoints and trends by augmenting the PELT algorithm (https://arxiv.org/pdf/1101.1438), with twin goals of spotting unusual activity in developer productivity and supporting dashboards for manual explorations of productivity. Validated system via simulations and used asymmetric Jaccard index to compare inferred changepoints and trend with the ground truth; asymmetry reflects greater danger of false negatives.

At Core Compute Analytics (CCA), aimed at classifying Borg compute jobs suitable for spatial shifting, with goal of reducing carbon and energy expenditure while limiting transmission costs, yielding a classifier that correctly recalls over 98% of spatially shiftable Google Compute Units (GCU). Furnished ML pipeline with real-time classification of previously-witnessed Borg jobs as well as novel jobs. Want to know more about spatial shifting at Google Cloud?Read more in The Economist: https://www.economist.com/technology-quarterly/2024/01/29/data-centres-improved-greatly-in-energy-efficiency-as-they-grew-massively-largerRead more here in Bloomberg: https://www.bloomberg.com/news/articles/2024-02-25/ai-increases-data-center-energy-use-google-pioneered-technique-could-help

Analyzed streaming providers (Netflix, Disney+, etc) using Python, PySpark, and 30TB of alternative to develop leading indicators, metrics, KPIs, and graphics for use in dashboards available to market analysts and portfolio managers, preparing traders for Netflix's 75% stock price decline from 2021 to 2022.

• As a member of CMC, used LASSO and other shrinkage estimators to optimize manufacturing processes for numerous important outcome variables. • Recommended experimental designs tailored to the needs of scientists, designs including full-factorial and fractional factorial designs, blocking.

• Built automated monitoring system for detecting anomalies and breakpoints in time series of credit card segments; co-wrote Python class for enhanced user experience• Conducted exploratory analysis of account holders with gig-economy jobs; evaluated their risk as borrowers and made recommendations for evaluating gig-economy income earners

As member of Advanced Analytics Team, used predictive algorithms to identify most promising four-year colleges for promoting BEXSERO® vaccine

Statistical Analytics InternExecuted months-long research project on identifying significant spikes in stock market time series data; conducted literature review on competing methods; performed high-N stochastic process simulations to validate techniques; worked on team that produced interactive graphics displaying results

Reported on international affairs for byline pieces for US news-wire service