A User Personalized and Adaptative Smart Home using Large Language Models

Smart home systems interconnect devices and allow occupants to control their environment for their activities. They also allow routines to be set up, automating actions on devices according to specific events. However, the configuration of routines and their further customization to changing needs are complex. This paper presents an advanced smart home system that leverages Large Language Models (LLMs) to significantly enhance home automation, addressing the complexity of customizing systems to individual user needs. An LLM suggests the smart home system's next best action at any event. A voice assistant listens to user feedback, allowing the LLM to dynamically update user preferences and adapt system behavior. Managing user preferences significantly improves the reactions of the system in a range of experimental scenarios. Experiments are carried out to evaluate several LLMs and distinct prompting styles. The results show significant difference in performance.