Navegando por Data de Publicação, começando com "2026-08-14"

Filtrar resultados por ano ou mês

Agora exibindo 1 - 1 de 1

A multi-agent layered architecture for recovering use cases flows in existing systems
(Universidade Federal de São Carlos, 2026-08-14) Carvalho, João Vítor de Oliveira; Camargo, Valter Vieira de; http://lattes.cnpq.br/6809743774407662; https://orcid.org/0000-0002-6439-4649; http://lattes.cnpq.br/5420793134490667; https://orcid.org/0009-0006-1776-2767; Wiese, Igor Scaliante; Valejo, Alan Demétrius Baria; http://lattes.cnpq.br/0447444423694007; http://lattes.cnpq.br/9546164790189830; https://orcid.org/0000-0001-9943-5570; https://orcid.org/0000-0002-9046-9499
Legacy systems frequently require reverse engineering activities. Among the various possible activities, understanding how specific use cases are internally executed by a software system stands out as an important task, requiring the identification of the software elements involved in their execution, such as classes, method calls, modules, and components. Although several studies have explored the use of artificial intelligence in activities related to software development and maintenance, no studies focusing specifically on the automated comprehension and extraction of use case execution flows in software systems were identified. In this context, multi-agent systems have emerged as a promising approach for solving complex problems composed of multiple stages and specialized skills, allowing agents to operate autonomously and collaboratively. This dissertation presents a tool-supported multi-agent approach for use case comprehension. The proposed approach/tool involves seven collaborative agents that work together to extract the software elements involved in the realization of a given use case. As a result, the tool automatically generates documentation composed of diagrams and textual descriptions that support software engineers in understanding the execution flow of the analyzed use case. Additionally, this dissertation contributes to the definition of the architecture of the developed tool, which is based on the integration of intelligent agents, MCP servers, and the A2A protocol, forming an infrastructure that can serve as a foundation for the development of new multi-agent systems. To evaluate the proposed approach, different instances of the tool, using different LLMs, were applied to the extraction of two use cases from a software system. The generated documentations was subsequently compared and analyzed by the author with prior knowledge of the analyzed software, producing satisfactory results regarding the comprehension of the extracted use cases.