Desenvolvimento de um terminal multi-cloud inteligente utilizando geração aumentada por recuperação (RAG) para operações de computação em nuvem

Abstract

This study presents the development of an intelligent terminal that employs the RetrievalAugmentedGeneration(RAG)techniquetosupportmulti-cloud environments, executing CommandLine Interface (CLI) commands extracted from the official documentation of Amazon Web Services (AWS) and Google Cloud Platform (GCP). The application, developed using the LLaMA model and run locally via Ollama, incorporated vector embeddings, semantic reranking based on Maximal Marginal Relevance (MMR), and the LangChain and Streamlit libraries to provide an accessible web interface. The prototype was tested by 42 participants who performed instance provisioning tasks in AWS and GCP environments. Their responses were evaluated according to f ive criteria: technical accuracy, completeness, clarity, usefulness, and relevance. The results indicated mostly positive evaluations, highlighting the system’s ability to provide contextualized responses, although limitations were observed regarding precision and error adaptation. The research demonstrated the feasibility of a locally executed RAG-based assistant, while also pointing to the need for more robust models and better-curated knowledge bases. The study concludes that the RAG approach has significant potential to transform the way users interact with multi-cloud systems, offering a foundation for future improvements in precision, explainability, and usability.