INSIGHT: Foundation-Model AI for Actionable Indoor Building Intelligence

Indoor environments remain one of the last blind spots in emergency response: responders routinely enter unfamiliar buildings with no usable map. Existing 3D building scans are too large to transmit, too costly to label, and lack a common vocabulary across agencies. This presentation introduces INSIGHT, an open-source, NIST PSCR-aligned research effort that applies foundation-model AI (Segment Anything Model 3) to camera and depth data to automatically identify safety-critical building features — exits, standpipes, AEDs, utility shutoffs — and encode them as ISO 19164-compliant 3D scene graphs. This approach compresses tens of gigabytes of raw scan data into megabytes, making complete indoor intelligence deliverable over FirstNet in seconds and filterable by role, so fire, EMS, and tactical personnel each see the information relevant to their operations. Attendees will see live segmentation results, role-tailored building views, and a path toward interoperable indoor intelligence for dispatch and field operations.