Guided Artificial Intelligence for Model-Based Systems Engineering through SysML v2

Model-Based Systems Engineering (MBSE) promises a single, consistent source of truth across the engineering lifecycle, yet its adoption at scale remains limited because contributing correctly to a system model requires an engineer to combine domain expertise, systems-engineering methodology, and modeling-tool proficiency at once. Large language models (LLMs) are increasingly explored to lower this barrier by generating SysML v2 content from natural language, but most existing approaches leave the model largely free to interpret a prompt and check correctness only after generation. This paper presents a guided-AI approach in which generation is constrained by an explicit, human-authored methodology container, and executed through a five-step agentic iterative loop with human confirmation. The approach couples a general-purpose LLM with a live SysML v2 model repository through a Model Context Protocol (MCP) server. We situate the approach within a broader target vision spanning document ingestion, AI-assisted model modification, and trade-off analysis. A preliminary, qualitative demonstration is reported on an agricultural-spraying unmanned aerial vehicle (UAV) case, including informal practitioner feedback. We discuss current limitations, distinguish explicitly between what is demonstrated and what remains prospective.