Michael Ellis

Walmart – Bentonville, AR

• Led development of a production causal inference library using DynamicDML (EconML), identifying ~$100M in incremental revenue opportunity through optimized sales-meeting targeting for Walmart Marketplace partners.
• Nearly doubled per-meeting ROI, enabling smarter sales prioritization across Walmart Marketplace partners.

The Home Depot – Remote – Bentonville, AR

• Technical lead, collaborating with cross-functional teams, to build voice and text based conversational AI systems capable of autonomous customer support, handling nearly 100 million contacts annually, leveraging LLMs, RAG, fine-tuning, NLP and NLU.
• Led the development of a RAG based frequently asked questions agent to improve the number of questions answerable through automation, leading to an A/B test that estimated a 4% increase in containable conversations.
• Designed and built an A/B testing framework-from hypothesis formulation and sample size determination to automated result analysis-allowing accurate causal inferences to drive strategy, innovation and product development. Enabled the first ever A/B test in a conversational experience for the company.
• Lead inventor on the Machine Learning-based User Intent Determination patent.

The Home Depot – Remote – Bentonville, AR

• Identified opportunity within the automated phone system for handling product rental inquiries, projected to automate 500k calls annually and estimated to save $5M per year by reducing transfers to contact centers and enhancing customer experience
• Spearheaded automatic speech recognition model fine-tuning and improvements, achieving a 45% increase in speech-to-text accuracy and $1.3M in annual cost savings.
• Developed a robust and accurate type-ahead system for contact center employees, improving agent text-entry efficiency.

Black Hills Energy – Fayetteville, AR

• Developed large-scale forecasting system for millions of time series to predict customer natural gas consumption to drive planning and business strategies. Improved forecast accuracy by 14%.
• Led development of data pipelines to prepare customer natural gas usage data and dashboard assets to analyze millions of customer consumption data. Partnered with NOAA to ingest over 2.2 billion weather data points to enhance our forecasting models.
• Deployed R Shiny web applications that allowed stakeholders to visualize and monitor forecasts and other important business metrics in real-time, cutting manual analysis time by 50%.

University of Arkansas – Fayetteville, AR

• Research: Computational Bayesian Methodology for massive high-dimensional datasets
• Advanced coursework: Probability Theory, Statistical Methods, Bayesian Data Analysis, Computational Statistics, Real Analysis, Measure Theory, Differential Equations

University of Arkansas – Fayetteville, AR

University of Arkansas – Fayetteville, AR

University of Arkansas – Fayetteville, AR

• Course covering an introduction to probability theory, hypothesis testing and regression analysis. Taught 2 course sections of approximately 40 students each semester.

University of Arkansas – Fayetteville, AR

• Taught 3 course sections of approximately 30 students each semester.

Student Course Evaluations

• "Michael is a great teacher who cares about his students and always tries to help. He thoroughly enjoys stats and helps you apply it to the real world."
• "Michael is very good at making difficult things understandable and he's also very concerned with how everyone is doing, which is awesome!"