LastSeen AI

LastSeen AI helps answer an everyday question: Where did I leave my keys? It turns observations from Meta glasses into a searchable memory of enrolled objects. The user captures a reference photo and names an item, such as a black car key. An Android app built with the Meta Wearables Device Access Toolkit streams the glasses camera and relays sampled frames to a Python backend. SAM 3.1 locates the enrolled object and hands. Muse Spark describes the object's visible placement relative to nearby landmarks. The pipeline checks for a resting placement and avoids duplicate updates before saving a sighting. The dashboard displays the saved image, placement description, and timestamp. Users type a natural-language question to retrieve the latest recorded location. The demo shows enrollment and two subsequent placements of the same key, with an updated lookup after moving it. The current prototype requires one-time manual enrollment. Automatic discovery and enrollment are future work. Inference is asynchronous, and processing waits are shortened in the demo. LastSeen explores a practical use for wearable AI: helping people recall where they put everyday belongings.