Integração de Modelos de Linguagem de Grande Escala com a técnica RAG para a Simplificação de Consultas a Manuais Automotivos

Abstract

The increasing complexity of automotive manuals poses a challenge for users, who often struggle to find specific information due to the length of these documents. In this context, Large Language Models (LLMs), such as GPT-4, emerge as tools to advance natural language understanding and generation, especially when integrated with the Retrieval-Augmented Generation (RAG) technique. Therefore, this article proposes an architecture designed to enhance accessibility and efficiency in retrieving vehicle information from automotive manuals, supporting multiple forms of input: text, audio, or image. The methodology employed involved a case study to analyze the responses and performance of the proposed architecture. It was observed that the architecture provides considerably fast and consistent responses, which demonstrates the effectiveness of LLMs integrated with RAG in enhancing access to information in automotive manuals. The results suggest that such an approach can simplify the user experience, reducing the time required to locate accurate information and increasing the reliability of the provided responses.