Architecting Scalable ML Platforms: The Integrated Infrastructure and Acceleration Behind Rovo

This post introduces ML Studio, Atlassian's enterprise-scale ML development platform, detailing its architecture and design principles for scalable, high-velocity, and governed ML systems. It highlights key capabilities such as reusable ML modules (versioned, sharable components with Git integration and automated CI/CD), a unified workflow orchestrator (CLI/portal for scheduling, composable workflows, hot clusters), and embedded compliance controls (user identity, domain-level, and column-level data access enforcement with tag propagation and default safety). The post also details improvements in developer productivity through automatic caching (saving 1000+ hours/month), experimental workflows (accounting for over half of ML Studio runs, saving 100+ hours/day), and easy cross-functional integrations with experiment tracking, feature stores, model registries, monitoring (ML Lens), and deployment platforms. Common workloads powered by ML Studio include GenAI, Predictive AI, and Training & Evaluation. AI features empowered include Rovo Search, Dev-AI, AIOps, Knowledge discovery, Confluence AI, and Loom AI/CoreML.