FocusVision is a real-time computer vision productivity tool that uses your webcam to detect and alert you the moment a distraction enters your environment — all without a single line of server code. It runs an object detection model (COCO-SSD) directly in the browser using TensorFlow.js, scanning your camera feed continuously and firing audio alerts and browser push notifications when it spots a phone, food, a second person in frame, a TV, or notices you've left your seat. Every session is tracked with a live focus score (% of time spent undistracted), a distraction counter with per-category breakdown, a longest uninterrupted streak timer, and a configurable goal (e.g. 25-minute Pomodoro). On session end, a full summary is generated and saved to session history (up to 30 sessions stored locally). A Ghost Mode hides the camera feed while keeping detection active — for users who find the live view distracting itself. Each distraction type can be toggled on or off independently, and alerts can be snoozed for 5 minutes. No API keys, no backend, no data ever leaves the device. The COCO-SSD model is loaded once and runs on the GPU via WebGL through TensorFlow.js. Tech: React · TensorFlow.js · COCO-SSD (Object Detection) · Web Audio API · Notifications API · WebRTC (getUserMedia) · Tailwind CSS · Vite