Pawception

Pawception is an AI assistant that helps owners watch over their pets and understand their routines using a single camera. Every pet has different habits, and every owner has different concerns. An owner might worry about furniture chewing, persistent barking, or a pet climbing somewhere unsafe. Other dangers—a kitten trapped in a can or smoke spreading through the home—may arise without anyone thinking to define a monitoring rule. Pawception combines personalized monitoring with open-vocabulary hazard detection. Owners describe events in everyday language. Prompt Polish uses Muse Spark to suggest clearer, fuller descriptions while keeping the owner in control. Smart Routing determines whether each event needs visual evidence, audio, or both; audio-only tasks send sound alone. In Online mode, webcam footage or playback-paced video is analyzed in windows. Muse Spark detects custom events and potential hazards, while a temporal engine merges continuing detections and applies configurable alert thresholds. Notifications link to supporting footage, and an interactive timeline shows activity, processing status, and surrounding context. In Offline mode, owners can search saved recordings for multiple events or explore their pet’s favourite places. SAM 3.1 segments the pet to build a 2D occupancy heatmap. Local MoGe depth estimation extends saved masks into approximate 3D hotspots within a reconstructed scene. Both views connect locations back to relevant video moments. We evaluated 54 clip/event decisions across eight development clips, achieving a mean positive temporal IoU of 0.832. The prototype brings personalized monitoring, proactive hazard awareness, and spatial insights into one reviewable workflow.