Nebius Group (NASDAQ: NBIS) announced the launch of Data Lab, a new workspace within its Token Factory platform designed to convert production logs and existing datasets into training data for AI model improvement workflows.

The platform allows teams to explore inference logs, curate datasets, and proceed directly to model iteration without rebuilding pipelines or transferring production data across different environments, according to the company's announcement.

Data Lab includes features for exploring inference logs and existing datasets in a unified interface, creating reusable datasets from production data without manual exports, and S3-compatible file connections that maintain data in existing storage systems without requiring raw data copies.

The workspace connects directly to post-training functions within Token Factory, enabling teams to move from data curation to model training within the same platform.

Nebius designed the system to address what it identifies as a workflow bottleneck where teams must export data from production systems, clean and reshape it through scripts, upload it to separate environments, and reconnect results back to deployment systems.

The platform supports S3-compatible storage connections, allowing enterprise teams to work with files in their existing storage systems through metadata connections rather than data transfers, according to the company.

Data Lab operates as part of Nebius Token Factory and focuses on enabling teams to use production logs and real user interactions as source material for subsequent model training cycles.