Agentic AI in IT Infrastructure Administration

Status: open / Type of Theses: Master theses / Location: Leipzig

Description of the current status and problems:

The management of modern IT infrastructure – ranging from on-premise servers and virtualized environments to containerized web services – has become increasingly complex. Administrators are required to interact with heterogeneous systems, APIs, configuration files, and extensive documentation. While recent advances in Large Language Models (LLMs) and agent-based AI systems show promise in automating reasoning and decision-making tasks, their systematic and secure integration into IT infrastructure management is still in its early stages. Current approaches are often tightly coupled to specific tools, and provide limited support for supervision, validation, and security of autonomous agent actions. In particular, the delegation of operational tasks to AI agents introduces new risks, such as unintended configuration changes, privilege misuse, or insufficient traceability of actions. There is a clear need for a generic, extensible framework that enables the controlled use of Agentic AI for infrastructure management while addressing governance, security, and validation concerns.

Tasks and research questions:

• Provide an overview of existing concepts, practical implementations and scientific publications in this context (see sources below as potential starting points)
• Design a generic, modular architecture for applying Agentic AI to IT infrastructure management
• Implement a proof-of-concept framework that supports pluggable components, such as:
  • access to APIs and system interfaces
  • integration of heterogeneous documentation sources for Retrieval-Augmented Generation (RAG)
  • AI agents capable of executing operational tasks
• Investigate how open-source AI frameworks and agent protocols (e.g., MCP and related approaches) can be integrated into such an architecture
• Develop and evaluate mechanisms for supervising, validating, and securing agent actions, including but not limited to human-in-the-loop concepts, policy enforcement, and auditing

Key research questions include:

• How can Agentic AI be integrated into IT infrastructure management in a modular and extensible manner?
• Which architectural patterns are suitable for safely delegating operational tasks to AI agents?
• How can security, transparency, and control over agent actions be ensured and evaluated?

Required qualification and knowledge:

• Solid background in computer science / data science
• Programming experience in Python
• Solid understanding of Machine Learning and Large Language Models
• Experience with Linux-based systems and IT infrastructure administration is helpful

Responsibilities within the scope of the work:

• Independent conceptual design of the framework architecture
• Implementation of a functional proof-of-concept
• Experimental evaluation of the framework and its security measures
• Documentation of design decisions, implementation details, and evaluation results
• Preparation and presentation of the results

Organization of supervision:

Supervision is provided by DSC ScaDS.AI (Dr. Robert Haase, Matthias Täschner). The results of the work are to be presented in a seminar, e.g., a master’s seminar at DSC ScaDS.AI.

Sources:

• https://doi.org/10.30534/ijatcse/2025/071422025
• https://www.mdpi.com/1424-8220/25/6/1666
• https://ieeexplore.ieee.org/document/10849561
• https://ieeexplore.ieee.org/document/11131150