Zoë Farmer

I build production-grade ML systems across a broad range of problem types — classification, forecasting, anomaly detection, and neural network inference. My focus is on models that are accurate, resilient, and fast: efficient data pipelines, careful hyperparameter optimization, and parallel training that holds up under real workloads. I also work across the broader data stack, from interactive visualization to agentic systems.

I studied Applied Math at the University of Colorado with a Computer Science minor and a passion to build real implementations for the topics I was studying in order to understand them at a deeper level. My first research job involved applying machine learning to helping improve experiment design in the university’s only applied math lab, ran by Mark Hoefer. There, I spent time building models to flag poor-quality lab photographs at scale and building massively parallel simulation frameworks for soliton gas dynamics in one-dimensional systems. That research grounding shaped how I approach modeling: start from what the data actually shows, and be honest about uncertainty.

Alongside my research work I helped start a small LLC with like-minded students through which we could pick up contracting work throughout the semester. After I graduated I found work on public health and global security at Talus Analytics, financial modeling and churn prediction at Red Dot Storage, and then landed at Square, where I’ve been since 2021. My work there spans production ML systems across classification, forecasting, and anomaly detection: marketing lead conversion ranking, hardware anomaly detection, web traffic classification. Beyond the models themselves, I care about the ecosystem around them — CI/CD pipelines, code quality, and tooling that lets teams move fast with confidence.

The work that excites me most sits at the intersection of rigor and creativity: finding the right framing for a modeling problem, designing a visualization that makes a complex output actually interpretable, or thinking through how an agentic system should handle uncertainty. I’m drawn to the full stack — from data pipelines and exploratory analysis through deployment and monitoring, and increasingly into LLMs, RAG, and agentic AI.

Outside work, I play Magic: the Gathering, make pottery, draw, read a lot of fantasy novels, and enjoy a good puzzle. I’m based in Toronto, Canada.