Teaching your webcam to spot the boss and hide your screen

What it does BossSensor watches your webcam feed and classifies faces in real time using a small neural network. If it decides your boss is approaching, the README claims it will hide your screen, though the exact behavior is unclear: the demo screenshots suggest a decoy editor window, while the text only says it “will monitor changes.”

The interesting bit This is office automation at its most petty. You curate two folders, data/boss and data/other, train a Keras model, and let OpenCV handle the live face grabbing. It is a charmingly over-engineered answer to a problem most people solve with Alt-Tab.

Key highlights

• Runs fully offline on a local webcam with no cloud APIs.
• Requires manual training with labeled images of your boss and non-bosses.
• Built on Python 3.5, Keras, OpenCV, and PyQt4; the README lists macOS as a requirement.
• MIT licensed, so the panic logic is yours to remix.

Caveats

• The README never explains the actual hiding mechanism; the demo flow implies a window swap, but the text only says it “will monitor changes.”
• The dependency stack is frozen in time: Python 3.5, PyQt4, and a manual Keras backend switch are all visible signs of age.
• You need what the README calls “lots of images of your boss,” with no guidance on how to collect them.

Verdict Better treated as a computer-vision learning exercise than a reliable workplace privacy tool, thanks to aging dependencies and mechanics that the README leaves unexplained.