About the New CPSA® – Advanced Level Module SWARC4AI: Software Architecture for AI

The iSAQB intro­duced the new Advanced Level module Software Architecture for AI (SWARC4AI) on December 19, 2025.

The SWARC4AI module intro­duces partic­i­pants to contem­porary software architecture concepts tailored for AI systems, enabling the design of powerful, scalable, and seamlessly integrable AI solutions. By the end of the module, partic­i­pants will have a solid under­standing of the key principles of software architecture for AI systems and will be equipped to apply them in the design and imple­men­tation of machine learning and gener­ative AI systems.

1. Has this new module been created now because of the great attention to LLMs and Diffusion Models, or was the topic on the to-do list anyway?

The “SWARC4AI” module covers modern software architecture concepts for designing powerful, scalable, and integrable AI solutions. It places a special focus on Machine Learning (ML) and Gener­ative AI systems, including Large Language Models (LLMs) and Diffusion Models. Given the current devel­op­ments in AI research and its broad appli­cation, it seems that these technologies are a central motivation for the devel­opment of the module. At the same time, the module is part of a broader offering for software archi­tects that is contin­u­ously evolving to meet the latest techno­logical challenges.

2. Where is the greatest need for learning and knowledge transfer in the field of AI system architecture? Is it about funda­mentals, i.e. designing meaningful systems in the first place, or rather about advanced concepts such as optimization, evalu­ation, and scaling?

The module covers a wide range of topics, including both funda­mentals and advanced concepts. It starts with an intro­duction to AI-relevant software architecture concepts and guides partic­i­pants through topics such as the design and devel­opment of AI systems, data management and quality charac­ter­istics in operation. There is a partic­u­larly strong focus on advanced topics such as scala­bility, security, compliance, robustness and inter­pretability. It aims to empower partic­i­pants in the funda­mentals and in the optimization and evalu­ation of complex systems.

3. Are we always talking about huge amounts of data, gigantic data centers, and millions in training costs? Or are there many examples of “small” and “local” AI?

The curriculum offers a holistic view of AI systems and empha­sizes that not all appli­ca­tions require gigantic amounts of data or large infrastructure. Topics such as embedded deploy­ments, resource-efficient approaches, and SaaS solutions show that there are many use cases for “small” and “local” AI. These are partic­u­larly suitable for scenarios in which energy consumption, costs, and data protection are crucial. At the same time, large-scale approaches such as LLMs and their integration into cloud environ­ments are also covered to represent the diversity of AI applications.

4. Do questions such as appro­pri­ateness, ethics and respon­sible use of technology (also: impact assessment) play a role in the curriculum? And ecology?

Absolutely, the module integrates these topics compre­hen­sively. It addresses ethical challenges such as bias, fairness, and trans­parency and intro­duces approaches to respon­sible AI and AI gover­nance. Inter­na­tional guide­lines such as the “EU Ethics Guide­lines for Trust­worthy AI” and other gover­nance documents are also taught. In addition, Green IT is a central component to promoting sustainable and resource-conserving approaches in the devel­opment and operation of AI systems. This shows that the curriculum considers techno­logical as well as social and ecological aspects.

5. How practice-oriented is the curriculum? Can I expect concrete help for the imple­men­tation of ideas?

The module is very practice-oriented. It offers exercises, case studies, and practical projects where partic­i­pants can apply what they have learned in real-life scenarios. Topics such as MLOps, design patterns, and the integration of ML models into existing systems ensure that partic­i­pants receive concrete tools and approaches for the imple­men­tation of ideas. The practice-oriented design makes it possible to address challenges in profes­sional practice directly.

6. Is security completely different for AI systems than tradi­tional IT systems?

Yes, AI systems have specific security requirements. The module covers threats such as adver­sarial attacks, data poisoning, and model inversion, which are less relevant to tradi­tional IT systems. Strategies for AI security are also presented, including measures for robust models, trans­parent devel­opment, and protection against attacks by explainable AI. These differ­ences make it clear that security concepts for AI systems must be tailored to the specific charac­ter­istics of ML models.

7. How much does the module require? Is it aimed only at people who have already designed and developed AI systems, or also at those who are just consid­ering it?

The module is aimed at people with basic knowledge of AI, machine learning, and data science. Experience with common frame­works such as TensorFlow or PyTorch as well as basic knowledge of ML methods such as super­vised learning and deep learning are required. However, it is not exclu­sively intended for experi­enced AI devel­opers, but also for archi­tects and engineers who want to deepen their knowledge in the design and integration of AI systems.

8. When can the first training courses be expected? Are there any announce­ments or empirical values as to when the first training courses could be certified?

The first version of the module will be available from 12/2024. The training courses should last at least three days and may vary depending on the provider. There is no specific information on the first certi­fi­ca­tions, but the training is part of the iSAQB’s Advanced Level Program and contributes credit points towards certification. It is, therefore, to be expected that providers will publish corre­sponding training courses in the near future.