Resume

My resume is available at: LinkedIn

Current

Data Scientist (Staff) at Meta.

I work on Content Understanding using Vision Language Models (VLMs), with a focus on evals for multimodal models that understand videos and images. I also worked on the router for Meta's LLMs, responsible for routing user queries to the best model and ensuring models have the necessary context to answer user queries.

Limited Partner. I invest in funds focused on software (typically SaaS), artificial intelligence, and robotics. For inquiries here, my email address is devon [.] brackbill [at] gmail [dot] com.

Founder. I have built several SaaS businesses. See more at PORTFOLIO.

Previous work

Data Scientist (Staff/Principal) at Pearl Health.

Pearl Health helps primary care physicians transition to value-based care arrangements. I build machine learning systems to identify highly actionable interventions to reduce health care costs and improve patient outcomes. For example, I built models to predict Preventable ED visits, and future Inpatient admissions for patients with chronic conditions. I ran these projects through all stages: from idea, to model creation, and finally to impact measurement. I also conducted research on Transitional Care Management to develop new products to reduce hospital readmissions.

Senior Data Scientist at Amazon Robotics.

I spent ~4 years at Amazon Robotics, where I built machine learning systems to optimize the flow of billions of items through Amazon's sortation centers and used discrete event simulation to develop algorithms and system optimizations. Concretely, I was the lead DS designing a novel robotic transportation system that increased throughput by 30%. I also worked on the Amazon Air Gateway and Air Hub projects, where I focused on the system design and integration of two tightly coupled systems: the robotic induction systems that place packages on robotic drives, and the robotic drives themselves. I developed metrics for these systems that were used across all Amazon Robotics, and I designed algorithms to optimally time the arrival of robotic drives at induction stations.

Data Scientist at Instagram for Content Moderation.

As a data scientist at Instagram, I worked on content moderation, specifically the measurement and automated removal of objectionable content that violates FB's community standards using ML systems. Here, I saw the importance of developing measurement systems that can demonstrate improvement in a production environment at scale--and not just showing impressive metrics on a golden data set. Measurement has been a particular focus of my career. The problems in content moderation lie at the intersection of free speech and creating safe online communities.

Data Scientist at Cooper Health.

I was a data scientist for a hospital in Camden, NJ working on population health. Our goal was to manage patient care to prevent them becoming sicker and incurring more expensive and intensive care. I built predictive models to forecast all-cause inpatient readmission risk, Ambulatory Care Sensitive Inpatient admissions, and other events to reduce unnecessary patient care. I also worked to integrate the Camden Health Information Exchange (HIE) into Cooper's Population Health workflow so that we could better manage patient care.

Education

PhD from UPenn. See more at PUBLICATIONS.

BA from McDaniel College.