Software engineers can deliver like never before. Even small, new teams can provision hardware, configure programs, and deliver applications at scale. This is awesome. Yet these tools require plenty of configuration and management. As such, our development pipelines can feel flimsy and overcomplicated, like a Rube Goldberg machine made of toothpicks. How can we get DevOps engineering to feel more like, well, engineering and less like…hacking at YAML until it works? This talk proposes we reframe operations tasks as development tasks using the language of platform engineering. Rather than a single technology, we propose a platform comprises all of the technologies a team uses to develop and ship applications. This talk will review operations work by Numerator’s deep learning team, and how framing this work in the context of platform engineering has helped communicate the priority of this work and led the team to more robust solutions. Attendees can expect some guidance on improving CI pipelines, some suggestions of working with infrastructure as code tools, and some honest (if a bit self-deprecating) humor.
Zach is a senior engineer at Numerator where he directs machine learning operations for the deep learning team. He's run the gamut of machine learning operations roles, from conducting and analyzing large-scale natural experiments as a data scientist at Civis Analytics to debugging networking code by hand-inspecting packets as a cloud engineer at a Chicago startup. He's originally from Fargo, North Dakota and holds a BA in Mathematics from Concordia College.