Ann Arbor, MI · eCommerce Personalization at Scale

Parker Moesta.

Senior Machine Learning Scientist · Domino's

I build ML systems that decide what millions of customers see. Then I prove the revenue.

See my work Résumé (PDF)

Selected Projects

My Work

I like building things end-to-end: custom deep learning architectures, recommender systems, computer vision, NLP, network analytics. Every card links to the source.

Milestones

How I Got Here

Economics, then enterprise tech, then production ML. The short version.

  1. 2016

    Economics → Enterprise Tech

    Graduated from the University of Michigan and joined Oracle in Boston. Led a chatbot proof-of-concept for a major retail partner that cut support-center volume by 25% — an early lesson in what shipping intelligence into a business actually takes.

  2. 2018–20

    The Pivot

    Returned to Michigan and rebuilt my toolkit from the ground up — Python, statistics, and the math underneath. Co-authored peer-reviewed simulation research (IEOM 2021) along the way.

  3. 2020–23

    First Production Recommender · Coupa Software

    Engineered an SVD collaborative-filtering system surfacing personalized product add-ons — contributing to $50M in incremental revenue — and prototyped a RAG-architecture Q&A assistant before "RAG" was a household term.

  4. 2024

    M.S. Data Science · University of Michigan

    Completed the master's (3.95 GPA) while working — deep learning research years spanning generative models, NLP, and computer vision. Most of the projects above were born here.

  5. 2023–24

    Experimentation at National Scale · Domino's

    Joined the Digital Experience team owning causal inference and A/B testing across national web and mobile channels. Shipped a deep-learning cart-abandonment model driving a 6% reduction, validated against holdout controls.

  6. 2025

    Building the Personalization Foundation

    Led the Customer-360 feature store — a distributed Spark platform turning 1B+ records into 400+ features powering 12+ production models — and built the company's first production recommender systems, attributed to $30M in annual incremental revenue. Podium finish at the Databricks Global GenAI Hackathon.

  7. 2026

    Senior Machine Learning Scientist

    Now owning applied ML and production architecture for eCommerce personalization end-to-end: candidate generation, ranking, low-latency serving, and the experimentation loops that prove what works — from research idea to system running at scale.

Parker Moesta outdoors on a tree-lined trail

About

Business problem first.
Research second. Production always.

I'm a Senior Machine Learning Scientist at Domino's, where I lead applied ML and production architecture for eCommerce personalization — the systems that decide, in real time, what millions of customers see. My work spans the full stack of modern ML: distributed feature engineering over billions of records, two-stage recommender systems, transformer-based sequence models, and the experimentation and causal-inference discipline to prove what actually moved the needle.

Detroit-born, Ann Arbor-based. I came to ML through economics and enterprise tech, and that path shaped how I operate. When I'm not shipping models, I'm in the mountains — or teaching a neural net to write recipes.

Now
Senior ML Scientist · Domino's
Focus
RecSys · Experimentation · Production ML
Education
M.S. Data Science & B.A. Economics · University of Michigan
Previously
Coupa Software · Oracle
Published
IEOM 2021 · Featured in a Databricks technical blog
Download résumé

Contact

Let's talk.

Email is fastest — I read everything.

parkermoesta@gmail.com
Neural Collaborative Filtering framework

Steam Game Recommender — Neural Collaborative Filtering

A personalized game recommendation system for Steam users built on Neural Collaborative Filtering, trained on user–item interactions with playtime as an implicit preference signal.

Live demo GitHub
Neural style transfer result

Neural Style Transfer

A Neural Style Transfer model built on the VGG19 architecture, trained to apply the style of one image to another while preserving the original content.

GitHub
Faces generated by the VAE

Novel Face Generation — Variational Autoencoder

A VAE trained on CelebA for 50 epochs that generates novel faces, visualizes the latent-space distribution, and manipulates generated images through latent arithmetic.

GitHub
WGAN-GP architecture

Conditional WGAN-GP for Animal Face Generation

A Conditional Wasserstein GAN with gradient penalty trained on an animal-faces dataset, generating class-conditioned novel animal faces.

GitHub
Grid of colorized historical images

pix2pix — Restoring Color to Historical Archives

Image-to-image translation with a UNet generator and PatchGAN discriminator, trained on paired historical/modern photos to colorize grayscale images of houses around Ann Arbor while preserving texture and detail.

GitHub
LSTM architecture

Novel Recipe Generator — LSTM

A Long Short-Term Memory network trained on a large recipe corpus that generates novel, creative recipes — a hands-on study in sequence generation.

Try the model GitHub
News word cloud

Concocting a BERT(iful) Soup — News Sentiment Analysis

A sentiment-analysis pipeline for climate-change coverage: NewsAPI for retrieval, NLTK preprocessing, a pretrained BERT model for sentiment, and Matplotlib/Plotly/WordCloud visualizations to surface trends across outlets.

Read on Medium GitHub
BERT classifier diagram

News Source Classifier — Fine-tuned BERT

A classifier that predicts an article's outlet (CNN, Fox News, MSNBC, Breitbart) from its text using a pretrained BERT backbone — 95.1% accuracy after 10 epochs.

Read on Medium GitHub
t-SNE embedding projection

Word2Vec + t-SNE — Semantic Geometry of the News

Trains Word2Vec embeddings on articles from different outlets and projects them with t-SNE to visually compare how sources encode meaning around the same topics.

Read on Medium GitHub
Animated growth of the PyTorch citation network over time

PyTorch Research Network Dynamics

Network analysis mapping the PyTorch research landscape: co-authorship communities, key figures, and a multi-level citation network tracing how the original PyTorch paper propagated through the field.

GitHub
CO2 change-rate chart

Climate Finance & the Decoupling of CO₂ from Growth

An econometric investigation of climate-related financial flows into developing countries: have these investments helped nations cut emissions while maintaining economic growth?

View the study
Interactive route map through the North Cascades

A Visual Journey Through the North Cascades

A five-day backpacking route rebuilt as an interactive map from Apple Health GPS exports and photo metadata — altitude color-coded, photos pinned to where they were taken.

Explore the map GitHub