Network Configuration with Natural Language

Abstract

This thesis presents a novel framework for leveraging LLMs to configure networks using natural language. Specifically, we introduce the use of an intermediate representation to facilitate precise network configuration, a comprehensive verification mechanism to ensure that configurations are correct, and an approach to context provision that facilitates scalable prompt-based learning. Our evaluation demonstrates that this solution is able to configure networks with high accuracy in real-world configuration tasks. Finally, we suggest directions for future refinement of our system.