Expert reasoning and preference data for domain-specific model alignment — 40+ qualified SMEs producing calibrated CoT and preference labels across STEM and finance.

Policy-driven task environments and failure taxonomies for enterprise agents — executable evaluation suites covering tool use, recovery behavior, and escalation accuracy.

Image, video, document, and text annotations for multimodal model training and evaluation — one calibrated pipeline across four modalities with cross-modal QA.

Synchronized sensor data workflows for embodied AI training and validation — teleoperation episodes with RGB-D, force, and kinematics aligned to a common clock.