MoeCode

I built Aegis around a simple question: how can you know that someone living alone needs help without constantly watching them through a camera? Aegis uses an ordinary camera to notice a possible fall locally using on device YOLO models. When something looks wrong, it checks a short event clip using SAM 3.1 and Muse Spark 1.3, then asks the person, “Are you okay?” They can answer by voice or button. If they ask for help or don’t answer, Aegis alerts their family. If they say they’re okay, it also checks for visible recovery before clearing the incident. I wanted the family to understand why they were being alerted. They can see what happened, what the models concluded, and whether the person responded. The app also includes household setup, privacy controls, event history, and a morning summary. One unexpected discovery became a big part of the project: Muse Spark could interpret movement from person masks encoded as hexadecimal text. I tested the same movement forwards, backwards, and frozen, and got different answers that reflected those changes. I built a comparison view so you can inspect what the image-based and text-only calls received and where they disagree. The demo runs recorded test footage through the actual app. The repository includes setup instructions, tests, and the saved experiments, including the mistakes. At the end of the day, this project is meant to give family members just a little more peace of mind over their elderly family members, or any family member for that matter, while maintaining a high level of privacy for the family members.