Integrating SysML V2 into a GRAG LLM Pipeline: Design, Implementation and Evaluation

Abstract

Translate Article

Systems engineering models grow in complexity, and stakeholders increasingly demand user-centric solutions. SysML v2's decentralized, API-accessible repository offers a promising bridge between formal engineering knowledge and non-technical users via natural language. However, meaningful stakeholder engagement remains difficult unless model content can be accessed intuitively and in real time. We present a proof-of-concept graph-retrieval-augmented-generation (GRAG) pipeline that treats a SysML v2 model as a semantic knowledge graph, thereby enabling the integration of valuable SysML v2 model data into the context of LLM prompts before inference. For this, we propose a GRAG pipeline design, its implementation, and a small-scale evaluation to assess the general feasibility. Our pipeline effectively augments prompts by querying relevant information, verifies that SysML v2 can serve as a knowledge graph in a GRAG architecture, and reliably enriches prompts with a high density of information. This research sets the foundation for use-case-specific integrations and for the use of a broad range of SysML v2 models.