University of California Los Angeles delivers an immersive theater experience with AWS generative AI services

This article describes how UCLA's Office of Advanced Research Computing (OARC) built an AI-powered system for an immersive theater production of the musical Xanadu, enabling real-time audience co-creation of digital media.

• Audience members drew sketches on mobile phones that were processed through AWS AI services and projected back as 2D images and 3D meshes
• System handled 80+ concurrent users with under 2-minute round-trip inference time using serverless architecture
• Deployed 24 SageMaker AI endpoints on EC2 G6 instances combined with Amazon Bedrock for managed model access
• Used Lambda, SQS, SNS, and EFS to build event-driven messaging pipeline for sketch processing
• Integrated multiple AI models: DeepSeek VLM, Stable Diffusion, SPAR3D, Nova Canvas, and Claude 3.5 Sonnet
• Implemented automated cost controls with nightly endpoint shutdowns; SageMaker represented 40% of cloud costs
• Deployed continuous integration via AWS CodeBuild with custom dashboard for human-in-the-loop control

UCLA successfully demonstrated that AWS serverless and managed services enable reliable real-time AI inference for immersive live entertainment experiences.