Mapeamento de modelos de processo de negócio em artefatos de especificação de software

Abstract

Software specification artifacts, especially UML diagrams, are often reduced or neglected in agile contexts due to the cost and time required for manual production, although they remain relevant for communication, alignment, and documentation across teams. In parallel, BPMN business process models are widely used in organizations and contain structured information (participants, activities, and flows) that can serve as a starting point to derive software specifications, bridging what the business describes and what the software must deliver. This work proposes a methodology to map BPMN models into software specification artifacts with the support of Artificial Intelligence (AI), applying mapping rules consolidated in the literature and adapting them into clear prompt-based instructions. The methodology is structured into four stages: selection and analysis of the baseline literature; systematization and adaptation of the rules; a case study using a BPMN model different from the original examples; and comparative analysis of the results. In the case study, prompts were applied to AI tools (ChatGPT, Gemini, and Claude) to generate a Use Case Diagram and its textual descriptions, a Class Diagram, and a Sequence Diagram, and the outputs were compared against a manual application of the rules using an objective checklist. A total of 50 runs (10 per model) were performed, and the overall ranking was: ChatGPT Plus > Claude Free > Gemini Pro > ChatGPT Free > Gemini Free. The main quality losses were not associated with straightforward rules, but with rules that require fine-grained traceability and global consistency across participants and flows. As a final outcome, the research provides recommendations on tools and prompting strategies for BPMN-to-UML mapping, considering efficiency, accuracy, and speed, with the goal of supporting companies and practitioners in the practical adoption of this approach.