Nowcasting Store Visits
Working for a retail client at dentsu X, I developed, implemented, and productionized a machine-learning-backed system for modeling actualized store visits prior to full realization. When an ad is delivered to a potential customer, there can be a delay before an in-store purchase. However, bidding optimization needs to occur at the point of ad delivery. The nowcasting system I developed estimates what full store actualization will be for an ad delivered now, cutting out the need to wait for actualization before optimizing bids. Enabled by this system, the marketing team was able to grow revenue for the client by 62% YoY.
Churn Measurement for B2B Retail
While at HD Supply, I developed a system for understanding churn in a retail setting. Unlike a subscription-based business, churn in retail can be ambiguous as a customer behaving normally and loyally may still be dormant for long stretches between purchases. Based on academic research, I developed and productionized a system for observing customer behavior and probabilistically estimating the likelihood that they have churned.
Measuring Innovation in Fine Art with Machine Vision
We developed a system for identifying innovative artworks over 300 years using machine vision. Pieces flagged as innovative were then classified as retro, mining older works for inspiration, or novel, pushing the boundaries of art in a completely new direction. We showed that the relative prevalence of retro and novel works in any given year came in fairly steady waves over time. Working with James Hale and Helias Antoniou, this project served as a wonderful intersection between my analytical and art historical education.
Waves of Inspiration: The Old and the New
Social Media-based Stock Market Prediction Project
Alongside a team, I developed a method for using social media sentiment to predict stock price volatility. We were particularly interested in the impact of each stock's community on that stock's price. We used unsupervised clustering and time-series methods to show a significant relationship between group social media sentiment and stock volatility. Stock sentiment scoring was provided by PsychSignal.
The Clustered Social Media Index in the Stock Market
Kaggle Competitions
I have been entering Kaggle competitions to improve myself as a data scientist and as part of my Applied Machine Learning coursework. I strive to use clean, intuitive, and cutting-edge machine learning techniques to find accurate and scalable solutions .
Fight Outcome Prediction Project
Using a Bayesian algorithm developed by Microsoft for online matchmaking, I developed a predictive model in R for mixed martial arts contests. I collected fight results using a web scraper I wrote in Python.
AT&T Advertising Claims
While working for BBDO-Atlanta, I developed and vetted advertising claims for AT&T Mobile. Using SAS Analytics and MySQL, my team and I used millions of data points describing network performance and statistical significance testing to allow AT&T to use statements like "The Nation's Largest 4G Network" in its marketing and advertising.