Leveraging Large Language Models for Generating Infrastructure as Code: Open and Closed Source Models and Approaches

Abstract

Infrastructure as Code (IaC) has emerged as a groundbreaking paradigm in the Software Engineering community and tech industry, offering unparalleled efficiency and scalability. By leveraging machinereadable code, IaC streamlines the management and provisioning of IT infrastructure, showing a way in a new era of automation, consistency, reproducibility, and error reduction across diverse environments. Despite its transformative potential, orchestrating IaC remains a labourious work, demanding a specialized skills and significant reduction in the manual efforts. Recognizing the importance for automation in today’s fast-paced industry landscape, this survey investigates the feasibility of harnessing Large Language Models (LLMs) to address the challenges in IaC orchestration. LLMs, a large neural networks based models, exhibit remarkable language processing capabilities and have demonstrated their efficiency in executing various instructions across a wide range of applications. Recently, they have also been adapted for code understanding and generation tasks successfully, which makes them a promising choice for the automatic generation of IaC configurations. The major focus of this thesis delves into the intricacies of IaC, exploring its utilization across different platforms and explaining the associated challenges. It examines the ability of LLMs in code-generation task, highlighting their potential significance in automating IaC workflows. Based on our experimental findings, we illustrate the importance of LLMs in the context of IaC and share insights of the possible applications in this domain. In conclusion, we highlight the challenges inherent in leveraging LLMs for IaC automation and emphasise the vast potential for future research endeavors to address these challenges that helps further in enhancing the efficacy of IaC orchestration and also to build an end-to-end system that comprehends the users intent and motivation.