Model-Based Systems Engineering with SysML is powerful but remains tool-heavy and difficult to adopt across company boundaries. This white paper presents the FAICE project's SysML Assistant — integrated into CATIA Magic Cyber Systems  Engineer — that converts natural-language prompts into valid SysML artefacts within a live model. We introduce a domain-specific benchmark measuring LLM  capabilities across SysML, MagicGrid, and conceptual SE knowledge, yielding  average F1 scores of 62–79 % for diagram generation with measurable gains from Retrieval-Augmented Generation. A user study with six engineers confirms  time savings in early modelling and identifies four requirements for productive use: persistent context, dialog-based clarification, transparent error handling, and schema-constrained output. A collaborative Mars Rover case study exposes cross-company friction points — profile dialects and abstraction-layer mismatches — from which we derive four practical guidelines  for LLM-enabled MBSE workflows. The paper closes with an agent-based reference  architecture and a roadmap toward SysML v2 migration, providing actionable recommendations  for deploying safe, interoperable AI assistance in multi-stakeholder engineering environments.

More information on FAICE