Personalized Home Assistant Using Large Language Model with Context-based Chain of Thought Reasoning
Abstract
Virtual assistant applications, e.g., that are utilized in a smart home context, rely on a combination of natural language processing (NLP) and commands, without specific understanding of user preferences in a flexible manner. There are limits to what can be learned by/ acted on by a virtual assistant based on a predefined trait that is decided a priori. This disclosure describes the use of a large language model (LLM) to enable a virtual assistant in a home context to have memory and the ability to learn and act on unstructured data about occupants in the home, their preferences, usage patterns, etc. along with information about the home itself and the devices within it, e.g., smart devices that can be accessed and/or controlled by the virtual assistant). The unstructured data is accessed with specific user permission and used for the purposes of serving users within the home. Chain of thought reasoning can be leveraged to decide when to take notes to update various user preferences and how to modify actions that are performed by the virtual assistant.
Creative Commons License
This work is licensed under a Creative Commons Attribution 4.0 License.
Recommended Citation
Cutbill, Adam; Monsler, Eric; and Hayashi, Eric, "Personalized Home Assistant Using Large Language Model with Context-based Chain of Thought Reasoning", Technical Disclosure Commons, (March 25, 2024)
https://www.tdcommons.org/dpubs_series/6814