← projects

Argus

view project →

Personal Project · Feb 2026

Home security systems are dumb — they record everything and flag nothing useful. Argus is an attempt to fix that.

How it works

Multiple camera feeds are ingested simultaneously using OpenCV. When motion is detected, frames are sampled and routed to an LLM which classifies the scene — distinguishing a person at the door from a raccoon knocking over a trash can — and fires an alert only when it’s actually warranted.

The goal was to keep latency low enough to be useful while avoiding the alert fatigue that makes most motion-detection systems useless in practice.

Stack

Python · OpenCV · LLM vision API