Masterarbeit
Automating Documentation Updates using LLM and Visualization of Cloud Infrastructure
Completion
2025/07
Research Area
Students
Oleksandr Boretskyi
Advisers
Abubaker Gaber
Prof. Dr.-Ing. Martin Gaedke
Description
Changing technological landscapes and the shift to cloud-based solutions, make it increasingly challenging to understand how processes work and where data is stored. The answers to these aspects should be clearly explained within the project's infrastructure documentation. Companies like CpX.Energy rely on Infrastructure as Code (IaC) approach to define and manage their cloud environments. Keeping documentation up to date with frequent changes is a manual, time-consuming, and error-prone process. Moreover, textual documentation alone does not always provide a clear understanding of complex cloud architectures, making visual representation through diagrams an essential component for improving infrastructure comprehension.
This thesis focuses on a two-fold solution: automating documentation updates through large language models (LLMs) and improving infrastructure understanding through appropriate visualization techniques. To achieve this, analysis of infrastructure code (specifically Terraform) using secure, internal LLM implementation for documentation and visual representation is required. The thesis aims to answer questions such as how to automatically identify infrastructure changes, transform them into human-readable documentation, and visualize complex cloud architectures for improved comprehension.
The objective of this master thesis is to find an approach or combination of approaches to solve the previously mentioned problems in the context of cloud infrastructure documentation and visualization. This particularly includes the state of the art regarding LLM implementations for technical documentation, secure AI deployment on internal infrastructure, and effective architecture visualization techniques. The demonstration of feasibility with an implementation prototype integrated into existing CI/CD pipelines is part of this thesis, as well as a suitable evaluation measuring both time efficiency gains and accuracy of automated documentation compared to human-generated equivalents.
