A practical accessibility problem for on-device AI
SIVO.GO, created by Loughborough graduate Gargi Agrawalla, won the 2026 UK James Dyson Award on September 18. The portable prototype alerts deaf and hard-of-hearing users to sounds such as fire alarms and doorbells through color, screen symbols, and adjustable vibration.
Its TinyML model was trained on more than 500 labelled audio samples and runs locally on an Arduino Nano 33 BLE Sense Rev2. A separate vibration sensor detects knocks and door movement because variable knocking was not reliable enough for AI classification.
The hybrid design is the important part
SIVO.GO does not try to classify every signal with one model. It uses local audio recognition where the pattern is clear and a physical sensor where context is harder to infer. That division keeps the device offline and avoids requiring installation in a hotel room.
Good edge AI is often less about making one model do everything and more about choosing where inference belongs.
The prototype still needs reliability testing, manufacturability work, user trials, and certification before it becomes a product. Its early design nevertheless shows how small, private, offline systems can solve a concrete problem without cloud dependence.