Agentes de IA para Automação na Seleção de Estudos em Revisões Sistemáticas da Literatura em Ciência da Computação

Abstract

The Systematic Literature Review (SLR) is an essential method in scientific research, but its screening stage, in which thousands of studies are manually assessed, is time-consuming and prone to inconsistencies. Solutions that automate this stage with Artificial Intelligence usually require programming skills, which limits their adoption. This work proposes and validates a modular architecture that automates the screening of primary studies, orchestrated by an AI agent on the low-code platform n8n, supported by a language model that runs locally, on the researcher’s own machine. The research follows the Design Science Research paradigm and validates the artifact through three case studies replicating existing reviews, each with a distinct profile of data source and information available for the decision. In the case with the broadest coverage, the system reached 98% recall over the recoverable studies, meaning it retrieved almost all relevant articles, and reduced manual screening effort by at least 28% and up to 91% in the scenarios evaluated, depending on the level of confidence adopted by the researcher. The results indicate that automating screening with AI agents on the n8n low-code platform itself is feasible and reduces manual work, provided that the researcher remains responsible for the final decision in ambiguous cases.