About

I build machine learning and optimization systems, and the experiments that show whether they’re working.

I came to this through statistics. I spent most of a decade in academic medicine designing studies and writing Bayesian software, then moved into industry: fraud models at Charter, then six years at Amazon Payments building systems that decide how real money gets spent. Since January 2026 I’ve worked independently, currently with NextEra Analytics on software for scheduling power plants as the grid makes room for more solar.

Work

  1. 2026 – now

    Software & ML Engineer · Independent (Mulvahill LLC)

    Embedded with NextEra Analytics, building a fleet-wide application for monitoring and optimizing power-plant scheduling, with the aim of making room for more solar and other renewables on the grid.

  2. 2020 – 2025

    Applied Scientist II · Amazon Payments

    • Designed and shipped a “predict and optimize” system that pairs ML forecasts with a linear program of over 1.2 billion decision variables to allocate customer incentive offers. It averaged a 20% increase in net profit, was adopted internationally, and now manages more than $250M in incentive spend.
    • Built a simulation and offline-evaluation tool on Ray, borrowing off-policy evaluation methods from reinforcement learning, so algorithm changes could be tested before they reached customers.
    • Designed a rotating, continual holdout experiment to measure the incremental value of Amazon Monthly Payments, a $2B zero-interest installment program.
    • Solved a cold-start problem in the payments ad-recommendation engine (more than a trillion impressions a year) and validated it with a global A/B test.
    • Was the org’s “bar raiser” for a company-wide causal inference model, reviewing 40+ use cases across 12 teams. Most recently, built agentic AI workflows that cut annual planning from 12 weeks of product-manager effort to 2.

  3. 2018 – 2019

    Principal Data Scientist · Charter Communications / Spectrum

    • Led data science for fraud detection: linear TV piracy and password sharing across Spectrum’s streaming apps and WiFi.
    • The password-sharing model and report landed in time to inform the Disney–Charter carriage negotiation, covering 22 channels and 15.8 million viewers (Wired wrote about it). Also built the model behind Spectrum Mobile’s primary profitability metric.

  4. 2010 – 2018

    Research Faculty & Biostatistician · University of Colorado Anschutz Medical Campus

    • Data manager, then biostatistician, then research faculty in Biostatistics & Informatics. Designed or analyzed more than 40 biomedical and public health studies and co-authored 15+ peer-reviewed papers.
    • Wrote open-source software for Bayesian models of pulsatile hormone data, taught reproducible research on campus, and set up version control and code review for a group of 15 consulting statisticians.

Projects

  • 2026

    reproducible-analysis-starter

    A project template for statistical analyses that stay reproducible and checkable when an AI assistant writes much of the code. Comes with a worked reanalysis of a public randomized trial.

    Repo Example report

  • 2016 – 2018

    pulsatile & libpulsatile

    An R package and C++ library for Bayesian deconvolution of pulsatile hormone time series using birth–death MCMC. Grew out of my master’s thesis.

    pulsatile libpulsatile

  • 2017

    multiMiR

    An R package for querying a large collection of predicted and validated microRNA–target interactions.

    Bioconductor

Background

2015
M.S. Biostatistics, University of Colorado Anschutz Medical Campus
2008
B.S. Sociology, University of Wisconsin–Madison
2018
rOpenSci Unconference invitee
2016
Maurice Davies Award for Excellence in Statistics
15+
Peer-reviewed publications, mostly clinical research. Full list on Google Scholar.