Justin Brantley Email and Phone Number

Developed a forecast blending framework that gathered disparate retail forecasts and automated the process of training ensemble models. Not only did this game-changing approach outperform existing forecast models, but it enabled peers to generate KPI predictions for entirely new products, timeframes, and geographies that hadn’t been previously studied.Led pilot effort to expand the Engineering Design Review process to include a model review, ensuring responsible data use & ethical ML/AI. Drafted standards & best practices for the Analytics & Data Science organization.Refactored legacy inventory positioning algorithms written in Python and R for redeployment to a production Kubernetes environment, adding comprehensive documentation and 100% unit test coverage yet reducing the size of the codebase by half. Integrated multiple data science products behind a genetic algorithm to find win-win options in the presence of competing business priorities.Authored "nordyverse", an internal R package used throughout the Analytics & Data Science organization to simplify connections to data sources, standardize workflows, communicate results, and promote best practices. CRAN-compliant with 100% test coverage; fully automated CI/CD pipeline.Collaborated with in-house Machine Learning Platform engineers to release V2 tools for data science at Nordstrom, integrating project templates, GitLab issue tracking, code review workflows, Docker images, CI/CD tools, and Kubernetes deployment patterns.

Produced retail affinity models to describe the product, brand, and color preferences of our customers. This feature set grew to support dozens of personalized experiences on Nordstrom.com, collectively worth tens of millions in demand annually.Modeled cost margins, allowing Nordstrom.com search results and product recommendations to optimize for the bottom line without sacrificing relevance. In 2020, this proved to be an invaluable tool that allowed us to strategically sell in-store inventory despite public health store closures.Untangled seemingly contradictory A/B test results to find a website refinement worth $4–6M in profit per year. Delivered production ML model in two sprints.Developed and continuously refined profitable algorithms to personalize search results, product recommendations, and editorial content shown on the Nordstrom.com homepage.Delivered a hybrid forecasting model capable of anticipating abrupt changes in retail demand (e.g. during a sale event), thus avoiding a common pitfall of conventional methods like ARIMA.

Reduced the time required to run A/B tests in half by developing a novel two-stage analysis technique for e-commerce data.Designed a Shiny web application to visualize the statistical power of an experiment in an intuitive way, making it easy for product teams to plan A/B tests and feature-rollouts with statistical rigor.Modeled retail return rates and customer visit frequency using time-to-event models, then derived bottom-line metrics that outperformed traditional KPIs. Introduced new metrics to business stakeholders, earning praise for my ability to make complex statistical concepts easy to understand.

Served as lead statistician (with faculty oversight) on analyses of clinical trials results; authorship includes articles published to The New England Journal of Medicine, The Lancet HIV, and JAIDS.Developed a new way to visualize complex time-series relationships by adapting algorithms designed for drawing metro (subway) maps; prepared such graphics for three peer-reviewed publications.

Created two R packages (one published on CRAN) to streamline ETL jobs and QA/QC checks, reducing the number of specimen-related data discrepancies by a factor of ten.

Configured the 21CFR-compliant specimen-tracking database for what has become one of the largest HIV biorepositories in the world. Led database trainings at clinical trial sites in Africa.