Inteligência artificial e automação na pesquisa científica
Rodrigo Fernandes dos Santos, Elisângela Cristina Aganette · RDBCI Revista Digital de Biblioteconomia e Ciência da Informação · 2025
Introduction: The growing volume of scientific publications in the field of Information Science has increased the demand for automated tools to support literature review processes. In this context, artificial intelligence (AI) agents based on large language models (LLMs) emerge as promising solutions to assist in reading and extracting information from academic texts.Objective: This study aims to propose and evaluate the use of an AI agent created by the authors for the semi-automated analysis of scientific articles, focusing on the identification of structural elements such as objectives, research gaps, methodologies, results, and future perspectives. Methodology: This is an applied research study, with a qualitative approach, exploratory design, and instrumental case study technique. A computational architecture was implemented using the libraries CrewAI, langchain_openai, and PyPDFLoader, allowing for autonomous reading of PDF files and systematic extraction of analytical information. The data were structured in YAML format, ensuring standardization and facilitating later analysis.Results: The agent correctly identified the structural elements of the articles and produced summaries that were compatible with human interpretations in most of the analyzed dimensions. However, it showed limitations in analytical depth and contextualization, highlighting the need for human mediation. Conclusion: The study demonstrates that AI agents can support systematic reviews by automating initial analysis stages. However, their use should be accompanied by qualified supervision to ensure the epistemological rigor of scientific interpretations. The proposed model represents a promising starting point for integrating AI into academic research workflows.