Lead Software Engineer on Nike Personalization Engineering team. I’ve been in software for 18 years, primarily in services the last 6. My previous experience was developing Medical Imaging Software for research and clinical trials. One of our studies was published in New England Journal of Medicine. My experience both in Science, and Engineering has allowed me to bridge the gap from Science to Engineering and bring ML models to market faster.
April 24, 2019 05:00 PM PT
Removing friction and automating the hard things, enabling Data Scientists to bring their Models to Production faster. I'd like to start with a small analogy. Before the cloud and deployment tooling, developers created services then often handed them off to an Operations Team to bring to production. This is where we currently are with bringing Machine Learning Models to production.
We know the benefits and cost savings of enabling developers to bring their services to production, so lets take that mindset to Machine Learning and Data Science. How we at Nike automated the Machine Learning pipeline, job scheduling, A/B testing and Serving. By reducing the effort and coordination with Engineering to bring Models to production, Data Scientists can develop more models, and rapidly test and measure improvements to existing models, all while removing the "throw it over the wall (of failure)" syndrome.
I'll be presenting concepts around automating the Machine Learning Pipeline (and dependency resolution) in an enterprise environment. Monitoring and Alarming of the Models at Processing time, automating A/B testing and how Databricks powers all this. This allows Data Science to own all aspects of a Model lifecycle, inception, development, productionalization, measuring and improving. I'll present the patterns and easy to use tools that allow for this.