Lifeline

After a flood, the urgent question is who is cut off and where a crew can get in. Lifeline takes before and after satellite photos plus a road network and finds which roads are cut and which buildings are isolated. It ranks where responders should enter first and writes a cited dispatch brief. Muse Spark judges flooding road by road, and disagreement is marked “uncertain” rather than guessed. An “Ask the map” agent answers typed or spoken questions (Muse Voice), and every citation is checked against the data. Tested on 20 SpaceNet 8 tiles, it flagged 638 of 639 truly cut-off buildings on 15 random heavily flooded tiles. Precision depends on how flooded the area is, and the limits are reported openly. Built toward “Push the Frontier”: using Meta’s models to understand and act on a physical system, a flooded road network, where responders need answers fast.