GENERATIVE ARTIFICIAL INTELLIGENCE: CONSTRUCTION AND EVALUATION OF A MODEL IN SUPPORT OF THE BRAZILIAN MARITIME AUTHORITY

This paper presents the development of a generative artificial intelligence model with recovery, expansion and generation architecture for the application of Brazilian Maritime Authority standards. The study addressed the construction of a vector base, based on selected norms, and the implementation of a tool using a large language model. The performance evaluation of the model used questions from a public tender of the Brazilian Navy for entry into the technical staff of waterway traffic safety, reaching 75% accuracy. Additional metrics such as fidelity, relevance of the response and context were also measured. The results achieved indicate robust performance of the tool, comparable to benchmarks of advanced models in the market. The work also discusses limitations such as hallucinations, biases and interpretability issues. Finally, it is concluded that the model represents a significant advance with the potential to automate consultations, streamline normative interpretations and improve staff training.