Iter8: AI-Driven Cloud-Native Release Engineering

What is it about?

DevOps, MLOps, and data-science teams face a basic question during each new production release of their applications or ML (Machine Learning) models: how to maximize release velocity, optimize business value, and protect end-user experience? Iter8 addresses this challenge by introducing the concept of metrics-driven experiments, an innovation that combines the industry's best practices for release engineering with bleeding-edge statistical learning algorithms for assessing apps and ML models and optimizing their rollouts in a principled manner. We made Iter8 available as open source here: https://iter8.tools

Featured Image

Photo by Caspar Camille Rubin on Unsplash

Photo by Caspar Camille Rubin on Unsplash

Why is it important?

Existing solutions and tools in the cloud-native community do not approach the problem of online experimentation (which we see as including SLO validation, conformance testing, A/B/n testing and rollouts, canary releases) with the required breadth and statistical rigor. Similar approaches adopted in the Web and mobile domains are not applicable to the unique challenges posed by cloud environments. Our novel problem formulation and underlying algorithms enable a wide variety of experiments, achieve high accuracy in identifying the best competing version of a cloud application or ML model, and can do so while maximizing business-reward metrics and satisfying SLOs.