Scaling AI Tutors: How Creatium Achieved 18x Faster Cold Starts with Cerebrium

Creatium has revolutionized online learning by creating immersive, AI-powered educational experiences that leverage pedagogical science and real-time AI systems to enhance productivity for instructional designers. However, as the platform evolved, infrastructure limitations became a significant challenge, causing provisioning delays and reliability issues. To address these hurdles, Creatium evaluated several infrastructure providers and ultimately chose Cerebrium for its ability to support both training and inference workloads efficiently. The implementation of Cerebrium's serverless infrastructure allowed for dynamic GPU provisioning, significantly reducing cold-start times to about 10 seconds and improving overall performance and reliability. This transition eliminated the need for prior workarounds, reduced operational complexity, and allowed engineers to focus on enhancing the learning product, all while achieving 99.999% uptime and cost efficiency.