Json-SysML: A Bidirectional JSON-Based Representation for SysML Models Enabling LLM-Augmented Model-Based Systems Engineering

Abstract

As Large Language Models (LLMs) are increasingly integrated into Model-Based Systems Engineering (MBSE) workflows, system model representation has become as a critical bottleneck. While the emerging SysML v2 introduces a native textual notation, SysML v1 remains the dominant industry standard, relying entirely on the XML Metadata Interchange (XMI) format for serialization. However, XMI’s verbosity and extensive non-semantic metadata make it inefficient for LLM-based processing, as it fragments semantic information across complex reference structures and rapidly exhausts context windows during iterative interactions. To address this limitation, we introduce Json-SysML, a compact, AI-optimized representation that encodes SysML v1 models into a standardized and semantically accessible JSON format. This schema is supported by an automated, bidirectional conversion pipeline that bridges Json-SysML and XMI-encoded SysML models, facilitating integration within existing MBSE environments..1 Experimental results demonstrate an average reduction in token count of over 97.6% compared to raw XMI, significantly improving suitability for LLM processing. Round-trip validation further confirms structural and semantic fidelity, verifying that Json-SysML models can be reconstructed and re-imported into conventional modeling tools. This approach establishes a concrete integration pathway between formal system modeling and data-driven AI methods, enabling more efficient interaction, analysis, and automation within AI-augmented MBSE workflows.