Case Study: Trajectory achieves 99.6% faster sandbox provisioning with Daytona

Trajectory speeds sandbox provisioning by 99.6% with Daytona

Trajectory, a machine learning company providing a continual learning platform, faced significant challenges in scaling their reinforcement learning training pipelines. Their self-managed Kubernetes infrastructure required 30 minutes to provision a single stateful sandbox, creating latency and limiting their ability to run thousands of parallel training cycles. This slow provisioning speed and lack of reliability under heavy workloads compromised their training efficiency and led to wasted compute resources. They evaluated several infrastructure providers before partnering with Daytona to solve their sandbox provisioning problem.

By integrating Daytona's sandbox infrastructure and utilizing its Snapshots feature for pre-configured environments, Trajectory achieved dramatic improvements. Daytona enabled them to provision stateful sandboxes in seconds instead of 30 minutes, a 99.6% reduction in time. This allowed Trajectory to reliably run over 700,000 sandboxes for long-running training tasks. The solution saved the engineering team over 15 hours monthly and provided the scalable, parallel infrastructure necessary for efficient reinforcement learning post-training, accelerating model iteration for their customers.

Trajectory