Junim: agente inteligente para conversão de sistemas Delphi para Java Springer

Abstract

The conversion of legacy systems developed in Delphi to modern ecosystems such as Java Spring represents a significant challenge in software engineering, characterized by high costs, operational risks, and the complexity of translating distinct architectural paradigms. This work presented the development of JUNIM, an artificial intelligence agent designed to automate and optimize this migration process. The methodology combined structural analysis of Delphi code—through the generation of a structured intermediate representation—with the transformational capabilities of Large Language Models (LLMs). The agent was orchestrated via a pipeline utilizing RetrievalAugmented Generation (RAG) to provide the LLM with factual context regarding the extraction of business logic from Visual Component Library (VCL) components and its mapping to Spring Beans, focusing on backend conversion and service-oriented architecture creation. The process also included an automated validation step with test generation and static analysis to ensure code quality and semantic preservation. The results demonstrated the feasibility of the AI-assisted approach in reducing manual effort and producing idiomatic, maintainable Java Spring code, establishing a robust methodology for the conversion of complex legacy systems.