The Artificial Intelligence Center at Novosibirsk State University (NSU) has introduced Kappa, a software framework designed to manage the full lifecycle of datasets for AI model training and benchmarking. The system handles data labeling, verification, versioning, publication, and model telemetry.
Kappa ensures data quality and reproducibility by tracking the authorship of every annotation. It allows for error correction within specific dataset versions, preventing recurring issues during training. The framework also supports benchmarking, enabling users to evaluate the performance of both proprietary and public AI models.
NSU is offering other universities two ways to access the platform. Institutions can use a free version hosted on NSU servers or purchase a license to install the framework within their own infrastructure. The annual license fee for local deployment is 100,000 ₽.
The platform has already been utilized in several projects, including a collaboration with the State Public Scientific and Technological Library of the Siberian Branch of the Russian Academy of Sciences. That project involved approximately 400,000 records and 100 annotators. Other applications include processing archival documents, facade documentation, and educational datasets.
