SmartEdge Assistant: A Voice-Controlled, Privacy-Preserving Home AI Hub

In the age of Internet of Things, smart devices from many manufacturers in different ecosystems are brought to the market to make our homes smarter. With the recent developments in AI, interacting with the home in natural language has become possible. We developed a prototype of a fully local voice assistant, operating without reliance on the cloud, to control and query a Home Assistant-based home. This local approach was chosen to ensure user privacy by keeping all interactions and data processing within the home environment. Furthermore, a benchmark, to evaluate such an assistant was defined and executed for the prototype using many open-weight LLMs. The benchmark results show that high-end consumer hardware is needed to run LLMs with around 30B parameters in order to produce acceptable results. To increase quality, we propose to refine the tools to control Home Assistant and provide richer descriptions of device capabilities, before increasing the model sizes.