Tag
The author details using AI coding assistants to build a custom smart-home tool that correlates audio recordings, sensor data, and sleep metrics to identify nighttime disturbances. The DIY project integrates a Raspberry Pi, microphones, Home Assistant, and a Garmin watch into a personalized web dashboard.
The paper introduces MIST, a synthetic dataset and framework for training multimodal voice assistants to control IoT devices in smart homes. It highlights significant performance gaps between open and closed-weight models in handling complex, speech-based tool-calling tasks.