AI-Powered Dance Archives: Arizona State University

The Arizona State University Artificial Intelligence Cloud Innovation Center (ASU AI CIC), powered by Amazon Web Services (AWS), has partnered with Arizona State University on a groundbreaking initiative to preserve and enrich a vast collection of rare dance documentation. This project leverages artificial intelligence and machine learning (AI/ML) to digitize and catalog more than 70 years of culturally significant dance traditions from around the world, making them accessible for future generations while honoring their cultural contexts.

The dance collection holds invaluable cultural heritage materials, but its potential was limited by the challenges of preservation and access. Manually cataloging decades of video footage is an incredibly time-consuming and resource-intensive process. This manual approach restricts the discovery of specific dance movements, cultural contexts, and historical details embedded within the archives. The university needed an automated, intelligent solution to not only digitize these materials but also to enrich them with detailed, searchable metadata while ensuring the ethical handling of all cultural content.

The CIC team built a fully serverless AI solution for automated dance video analysis on AWS, structured around four main components:

• AI/ML Video Analysis: Amazon Bedrock Nova Pro powers multimodal video segmentation and movement classification, and ethnochoreographic annotation using a two-phase pipeline. Retrieval-augmented methods ground metadata in archival context, enabling rich, standardized classification across 40+ fields.

   

• Content Ingestion: Amazon S3 serves as the staging area for secure video upload and storage. Automated workflows, via AWS Lambda, process uploads and trigger video analysis. Quality checks and video segmentation are handled serverlessly for scale and reliability.

   

• Frontend: The user interface is built in React and TypeScript, providing a drag-and-drop web app experience for uploading videos, monitoring progress, and reviewing analytical results. Deployment is managed via AWS Amplify, ensuring accessibility and a seamless user experience.
• Infrastructure: API Gateway exposes REST endpoints for job creation and status, while AWS Lambda delivers serverless backend orchestration. Amazon DynamoDB manages metadata, job status, and results persistence. All resources are provisioned and maintained through AWS CDK for rapid, reproducible deployment and maximum cost-efficiency.

This architecture transforms traditional manual cataloging into a fast, scalable digital workflow, reducing analysis time from hours to minutes and making dance research more accessible and consistent.

• AI/ML Video Analysis & RAG Architecture: AWS Bedrock models are used for automated video analysis, identifying dance movements, styles, and cultural origins using frameworks like Laban Movement Analysis (LMA). Retrieval-Augmented Generation (RAG) grounds automated metadata in archived content.
• Content Ingestion: Amazon S3 serves as the staging area for video uploads (batch upload up to 10 videos, each 10 minutes to 4 hours). Automated workflows handle format conversion, quality checks, and video segmentation for effective AI processing.
• Frontend: The web-based user interface is designed for efficient batch uploads and simplified management. Built with modern frameworks and hosted on AWS Amplify, it ensures accessible batch uploading and results review.
• Infrastructure: The solution uses AWS Lambda for serverless compute, Amazon S3 for secure storage, and DynamoDB for cataloging output and managing metadata. All AWS resources are provisioned and managed through AWS CDK to maximize scalability, cost efficiency, and rapid deployment.

This automated approach accelerates cataloging, standardizes metadata, and transforms a previously manual process into a better digital workflow.

This project redefines how cultural institutions can approach the preservation and study of performance-based heritage. By applying AI to archival processes, the solution drastically reduces the time required to catalog historical materials, making once-hidden content searchable and accessible to researchers, educators, and the public.

For the first time, scholars will be able to perform granular searches for specific dance movements, compare traditions across different cultures, and analyze the evolution of dance forms over time. This initiative not only safeguards irreplaceable records of human expression but also creates new avenues for academic research and cultural appreciation. The successful implementation of this AI-powered system will serve as a powerful model for other archives, museums, and heritage organizations worldwide.

“Working on this prototype shifted how I think about AI in the archive. I no longer see it as a tool for automation alone, but as a way to surface patterns across decades of embodied knowledge that would otherwise remain hidden. The key is ensuring that technological efficiency never overrides cultural responsibility.”
 Shan Chuah, Fine Arts Specialist, Cross-Cultural Dance Resources Collections at ASU

The framework developed for the ASU Dance Archive has significant potential for wider application across various sectors.

• Cultural Heritage and Museums: Other archives holding video or audio collections (e.g., oral histories, musical performances, historical events) can adapt this architecture to automate cataloging and enhance searchability.
• Athletics and Sports Science: The video analysis models could be repurposed to study athletic movements, providing insights for training, performance analysis, and injury prevention.
• Performing Arts Education: Educational institutions could use this technology to create interactive learning tools, allowing students to analyze and compare different performance techniques in detail.
• Media and Entertainment: Media companies could leverage a similar system to automatically tag and catalog vast video libraries, improving content discovery and management.

By demonstrating how AI can ethically and effectively handle sensitive cultural data, this solution provides a blueprint for any organization seeking to unlock the value hidden within its archival media.