Integrating LLM-based modeling assistants into Capella: experiments and architectural tradeoffs

View at publisher Summarise with VIRAT
Authors: Tiago Cruz, Margarida Dias-Amaro-Morato-Barata, Tomás Farinha-Nascimento, Iulian Ober, Mario Plagaro-Benito
Cited by: 0
References: 13

Summary

No summary yet. VIRAT can read this paper and write a plain-language summary of its aim, methods, key findings and limitations.

Summarise with VIRAT Free to try with your Science Catalogs account.

Abstract

The rise of large language models (LLMs) opens new opportunities for supporting model-based systems engineering (MBSE). This is particularly true in phases that rely heavily on text, such as requirement capture, or on idea generation and conceptual analyses, such as Operational Analysis (OA) in Arcadia/Capella. In this work, we report on a set of experiments and prototypes exploring the feasibility of LLM-based assistance in MBSE. First, we defined evaluation criteria and tested a set of different LLMs for their suitability in OA tasks, highlighting strengths and limitations across models. Second, we prototyped four integration approaches with Capella: a lightweight assistant that generates tables of elements in a text format (CSV) to be imported in a Capella model, a direct code generation approach where the LLM produces Python scripts for model manipulation, a LangGraph-based assistant that implements the ReAct paradigm with structured tool calling and memory management to enable iterative reasoning over model elements, and an integration approach relying on the Model Context Protocol (MCP), in which we provide a set of MCP tools for editing the Capella model that any MCP-compatible LLM assistant can readily use. We discuss the trade-offs of these approaches in terms of usability, robustness, and extensibility. Our findings suggest that LLMs can meaningfully assist in MBSE tasks; in particular in OA, and potentially in other Arcadia phases, but that integration choices significantly shape the capabilities of the assistant. We conclude by outlining challenges and opportunities of embedding LLM-based assistants into MBSE workflows.

Fields of Study & Subjects

Systems Engineering Methodologies and Applications Model-Driven Software Engineering Techniques Safety Systems Engineering in Autonomy

Publication details

Pages
1–8
Publication date
14 January 2026
Language
EN