Examines whether microcontrollers and TinyML can detect aggressive door slamming as an early warning signal for domestic violence.

Uses LSTM autoencoders with surrogate models and SHAP explanations to interpret anomaly detection decisions on IoT sensor data.

Combines pre-trained vision models with edge processing to flag unusual activity for smart city deployments in farms, castles, car parks, and bus stops.

Develops distributed analytics that can move between edge, fog, and cloud nodes so hygiene services operate reliably without constant connectivity.

Demonstrates how semantic web technologies running on openHAB can wrangle personal IoT data at the edge, reducing unnecessary data transfers for smart-home data science.

Uses proximity between family members to adapt smart-home permissions, ensuring children access sensitive content only when guardians are nearby.

Generates realistic smart-home activity data by scripting scenarios through a GUI and publishing them to openHAB via MQTT so researchers can prototype analytics and interfaces without installing hardware.

Demonstrated seamful design techniques for BLE-based indoor experiences through the Ghost Detector museum game.