AI and Authentic Comms
October 18, 6:00pm PT, Jarvis Room
It's not uncommon, you have a problem where you finally need machine learning. You've got a great data science team. They come up with something, and put it out there, and then 6 month's later it's a broken down mess. We love our data science friends, but it's not their job to build great production software. As engineers, it's ours. In this talk, we'll go through the story of DigitalOcean's fraud prevention tooling, lovingly dubbed "sketchy". You'll learn about it's origin, our first attempt at using machine learning, and how we've come up with practical production machine learning strategies. You'll find out more about "standard" site reliability techniques, as well as how we change some of them for our ML work. This talk includes practical tips for everyone, but is best for those who have some experience operating production software.
Check out all of our other sessions.
Check out all of our other speakers.