Towards LLM-Powered Consistency in Model-Based Low-Code Platforms

Low-code platforms often use various models that define the application built by citizen developers. With the increasing size and complexity of the applications built using low-code platforms, the number of required models and the dependencies between them expand. However, with increased complexity,...

Ausführliche Beschreibung

Gespeichert in:
Bibliographische Detailangaben
Veröffentlicht in:International Conference on Software Architecture Companion (Online) S. 364 - 369
Hauptverfasser: Hagel, Nathan, Hili, Nicolas, Bartel, Alexander, Koziolek, Anne
Format: Tagungsbericht
Sprache:Englisch
Veröffentlicht: IEEE 31.03.2025
Schlagworte:
Artificial intelligence
Complexity theory
Computer architecture
consistency
DSL
LLM
low-code development platform
meta-model
Model driven engineering
Prototypes
Software architecture
ISSN:2768-4288
Online-Zugang:Volltext
Tags: Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
Beschreibung
Zusammenfassung:Low-code platforms often use various models that define the application built by citizen developers. With the increasing size and complexity of the applications built using low-code platforms, the number of required models and the dependencies between them expand. However, with increased complexity, keeping these models consistent during the development or evolution of the application is crucial and often a non-trivial task for citizen developers. In this paper, we present four approaches to how LLMs can be used and integrated on the architecture level into low-code platforms to (i) create consistent models automatically, (ii) keep models consistent based on different types of dependencies, and (iii) support users in maintaining consistent models during the development. We implemented the approaches in a prototype and evaluated them in an exploratory study. The results show that state-of-the-art LLMs are capable of preserving consistency for low-code models as well as generating correct and consistent models in various scenarios.
ISSN:2768-4288
DOI:10.1109/ICSA-C65153.2025.00058