In recent years, we have gained an essential insight about the field of software development: DevOps is no longer a nice to have – it is absolutely necessary. A fast pipeline of Continuous Integration, Continuous Delivery, and Continuous Deployment delivers value to customers.

Delivering value and solving problems is also the goal of every machine learning model. However, building the model is the easy part. The real challenge is to build an integrated machine learning system.

You’ll leave this talk with an understanding of how we can apply learnings from "traditional" software engineering in a data science environment. You’ll learn how we can version, test, and monitor our model, our data, and all the other moving parts of our ML system. We will talk about different degrees of maturity in MLOps, the big picture of pipeline architectures, and the nitty-gritty details about why you don’t want to deploy your Jupyter notebooks to production. Also, we will explore weapons of choice for different parts of your ML system and common misconceptions about machine learning in production.