Reliable Cloud Operations Using Transformers

Managing large-scale public clouds presents significant challenges, particularly in analyzing the risk of commands executed by operators to guarantee that they do not cause damages to the infrastructure. Traditional rule-based systems have been used but fall short in scalability when managing operations at a global scale. Although machine learning techniques have been proposed as alternatives, they have typically been applied in small-scale environments. This article presents a novel approach to address the complexities of global-scale command risk analysis and standard operating procedure verification using large language models. Through an extensive evaluation in production, the technique shows more suitability than existing methods.