University of Twente Student Theses
CLAIR : Generating On-Demand Low-Code Application Documentation through Knowledge Graph and LLM-based Multi-Agent System Integration
Eichhorn, Tim (2025) CLAIR : Generating On-Demand Low-Code Application Documentation through Knowledge Graph and LLM-based Multi-Agent System Integration.
PDF
11MB |
Abstract: | Low-Code has revolutionised software development by enabling rapid application creation with minimal coding effort. However, as Low-Code applications scale, challenges related to documentation, maintainability, and technical debt become increasingly prevalent. Inadequate documentation impedes collaboration, maintenance, troubleshooting, and knowledge retention, particularly in agile development environments where documentation is often disregarded. This thesis introduces CLAIR (Connecting Low-Code and Artificial Intelligence for RAG), an AI-driven documentation assistant that leverages a knowledge graph and a LLM-based Multi Agent System to generate on-demand, context-aware documentation for Low-Code applications. The tool is validated using the Mendix platform, and the study employs a Design Science Methodology to design, develop, and validate the proposed solution. A comprehensive literature review explores challenges in Low-Code documentation, emphasising issues such as fragmented knowledge, poor traceability, and the impact of missing documentation. A survey and case study at CAPE Groep, an Low-Code consultancy firm, further highlight the documentation needs of Low-Code developers and business analysts. To address these issues, CLAIR integrates knowledge graphs to structure and store Low-Code application data, enabling efficient querying and multi-hop reasoning. Additionally, a Multi-Agent LLM System dynamically generates and enhances documentation based on application data and user queries. CLAIR automates documentation generation across various phases of the Low-Code Development Lifecycle, including the design, development, and maintenance phases. Key features include automated extraction of domain models, microflows, and dependencies, generation of high-level summaries and technical details, and support for troubleshooting. The system enhances maintainability, knowledge retention, and team collaboration by ensuring up-to-date, structured, and queryable on-demand documentation. Validation was conducted through expert evaluations and a series of test cases using Technical Action Research, demonstrating CLAIR's ability to generate accurate, usable, and context-aware documentation. Findings indicate that automated documentation significantly reduces the time and effort needed to create high-quality documentation. This documentation leads to reduced cognitive load, technical debt, and maintenance effort, making it a valuable asset for Low-Code development teams. This research contributes to the fields of Low-Code development, automated documentation, and AI-driven knowledge management, proposing an innovative approach that combines knowledge graphs and LLMs to enhance documentation processes. By bridging the gap between Low-Code application development and AI-driven automation, CLAIR sets a foundation for future advancements in intelligent Low-Code documentation and maintainability solutions. |
Item Type: | Essay (Master) |
Clients: | CAPE Groep |
Faculty: | EEMCS: Electrical Engineering, Mathematics and Computer Science |
Subject: | 50 technical science in general, 54 computer science |
Programme: | Business Information Technology MSc (60025) |
Link to this item: | https://purl.utwente.nl/essays/106037 |
Export this item as: | BibTeX EndNote HTML Citation Reference Manager |
Repository Staff Only: item control page