We will present a reference architecture for integrating a deep learning model into a production web application backed by containerized micro-services. The use case is a recommendation engine that uses collaborative-filtering to predict which items a user will be interested in. Starting with an overview of the data-science lifecycle, we will quickly focus in on the run-time architectural components. We will dive into all of the components in the flow including: gathering training data, training the model, saving the new model, serving the model via a web service, scoring new items as they are added, and finally producing a top K recommended items for a user. We will also review the CI/CD pipeline used to build and deploy changes and discuss how additional features would be added. Finally, we will review our definition of what it means to be "production-grade" and discuss how all of these criteria were achieved.
Chris is a hands-on software architect, consulting adviser and thought leader at J. B. Hunt Transport. Chris's passions include performance tuning, security, application portfolio rationalization and integration. You can usually find him at the whiteboard leading a discussion, or pouring over logs and providing performance analysis. Chris is currently working on enterprise architecture emerging technology for the J. B. Hunt 360 cloud platform.