An agent-based approach for the automatic generation of valid SysMLv2 Models in industrial contexts

This is the authors' abstract. We don't add key points for this paper.

Automating the generation of valid SysML v2 models from natural language specifications holds promise for advancing Model-Based Systems Engineering (MBSE) in industrial settings. However, current approaches based solely on Large Language Models (LLMs) often fail to meet the syntactic and semantic rigor required by formal modeling languages. This paper introduces a domain-informed, agent-based framework that combines LLMs with structured retrieval and iterative validation to synthesize correct SysML v2 models. The system integrates Retrieval-Augmented Generation (RAG) using a curated repository of SysML v2 examples and enforces compliance through a validation engine based on the official ANTLR grammar. Experimental results across diverse MBSE scenarios demonstrate that the integration of retrieval and validation mechanisms leads to a substantial improvement in model correctness and semantic alignment, beyond what each component achieves individually. This combined effect enables reliable, closed-loop generation of formal models from natural language, illustrating how domain-specific integration can transform general-purpose LLMs into reliable assistants for engineering design tasks.