Utilizando LLM function calling para a construção de assistente virtual

Abstract

Large Language Models (LLMs) have excelled in solving various types of problems, especially in the context of Natural Language Processing (NLP). This capability has been widely explored through the development of a functionality known as Function Calling, which enables LLMs to interact with external tools—such as APIs and functions—via natural language commands, acting as a bridge between textual understanding and real-world actions. In this context, this study aimed to explore the construction of bots integrated with LLMs through the use of Function Calling. To this end, a bibliographic and documentary review was conducted on the fundamentals of LLMs, the concepts related to Function Calling, its use in computational contexts, and its relation to Deep Learning principles applied to NLP. In addition, tools such as LangChain were analyzed, which facilitate the development of intelligent bots capable of integrating LLMs with external APIs. As part of the practical investigation, a generic bot was developed using Google’s Gemini model, integrated with a simulated REST API, featuring functionalities such as attendance registration and listing of calls. The tests demonstrated that the adopted approach was effective in simulating the operation of the Function Calling feature in LLMs, highlighting how these models can interact with APIs, interpret natural language commands, and generate automated responses based on the returned information. Furthermore, the integration of this technique contributed to a deeper understanding of the limitations and potential of this approach, serving as a foundation for future research and applications in the field of artificial intelligence applied to natural language understanding.