Design of a Quality Management System for AI

This technical report discusses the design of a Quality Management System (QMS) based on the European Union's Artificial Intelligence Act (EU AIA). The EU AIA mandates that providers of high-risk AI systems establish a QMS to identify, analyze, evaluate, and mitigate risks, ensuring compliance with training, validation, and testing data. The report outlines the current state of research, which primarily focuses on conceptual frameworks for AI risk assessment, while often neglecting practical tools for human involvement in the documentation process. To address this gap, the authors propose a generic design and architecture for a QMS that includes legal requirements and a prototype implementation. The prototype integrates large language models (LLMs) as examples of AI systems and emphasizes risk management. The report evaluates the prototype through qualitative assessments based on user scenarios and technical evaluations of storage and performance requirements. The findings aim to provide a comprehensive approach to compliance management for AI systems under the EU AIA.