Leveraging Large Language Models for Intent-Based Networking: A Framework for Automated Configuration and Optimization

Intent-Based Networking (IBN) enables network operators to specify desired network outcomes at a high level without detailing the procedural steps required to achieve them. However, translating these intents into accurate, executable configurations remains complex and error-prone, especially with modern networks’ growing scale and complexity. This paper presents a framework that leverages Large Language Models (LLMs) to automate the generation of Kubernetes YAML configuration files from natural language prompts, effectively bridging the gap between operator intent and precise configurations. Our approach introduces a novel Graph Neural Network (GNN)-based ranking algorithm to generate metadata from unlabelled YAML files, which, combined with fine-tuning techniques, empowers LLMs to interpret operator goals and produce syntactically and semantically correct YAML configurations. Experimental results demonstrate that the framework achieves high accuracy in term relevance ranking (72%) and efficiently generates YAML configurations, significantly improving operational efficiency and reducing human error in scalable IBN systems for future networks.