We’re building a single AI model that understands objects, spaces, and their relationships - from simple video inputs, without manual labels.