Towards Model Compliance Using Generative Agents: A NetLogo to Sequence Diagrams Experiment

Read full text Summarise with VIRAT
Authors: Benoît Ries, Nicolas Guelfi, Tiago Sousa
Cited by: 0
References: 12

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

Simulation helps software engineering teams explore complex system behavior, yet NetLogo code remains hard to interpret without design-level documentation.Reverse engineering using generative agents offers a solution to reconstruct models that visually document design.This paper reports on an experiment using generative AI to reverse engineer NetLogo simulations into restricted UML sequence diagrams that represent execution scenarios, produced by specialized generative agents.To ensure compliant model generation, we orchestrate the generative AI task in multiple specialized steps with intermediate compliance audits, each step guided by personas and domain-specific rules.We evaluate the approach on ten public NetLogo simulations paired with eight generative AI models, for a total of 80 experimental runs.Results show that Gemini 2.5 Flash achieves the best results, followed by GPT-5-mini, GPT-5, Devstral; Qwen3, Maverick remain promising, whereas GPT-5-nano underperforms.The experiment shows that orchestrating generative agents with iterative compliance audits improves model compliance.

Fields of Study & Subjects

Model-Driven Software Engineering Techniques Software Engineering Research Advanced Software Engineering Methodologies

Publication details

Journal / source
International Conference on Model-Based Software and Systems Engineering (MODELSWARD)
Pages
442–451
Publication date
1 January 2026
Language
EN