My dissertation covers 3 main products of my work.
These are two designs called Scalable AI Accelerator, short SAIA, and Scalable Kubernetes Inference Platform, short SKIP, as well as a design method for system of systems design.
The two designs both provide an approach for an AI inference platform that operates on top of an existing HPC environment and a private cloud such that even sensitive inference workloads can be processed.
Moreover, both platforms are designed to be highly reliable and robust with support for redundant deployments and even geo-redundancy to achieve active-active failover between multiple sites.
It should also be noted that we have already published about SAIA [2].

The approach for creating both designs is a method I introduced in my work specifically designed for system of systems design in research.
I have extended the research method for system design [3] into the Iterative Research Method for System of Systems Design, short IRMSSD.
This method allows for splitting requirements across a number of subsystems, which in turn also might consist of subsequent systems and then evaluating these subsystems before the systems that depend on them are evaluated.
IRMSSD is explained in detail in the methodology chapter of my dissertation.

It should also be noted that many designs that I propose in my dissertation have only been partially or not at all been evaluated yet.
As both designs are quite extensive, this was not feasible in the 3 years I was a PhD student.
Moreover, exploring the designs requires resources that are not necessarily available, especially as the two designs achieve the same features but with different software stacks and different limitations.
It is therefore more feasible that only one of the two designs, SAIA, is going to be implemented.
The details on the designs can be found in the dissertation itself and the status of the implementation for SAIA, in the respective paper [2].

While my dissertation stands for itself, I want to mention a few points about how I wrote it.
With 300 pages, 26 tables and 44 figures my dissertation is quite extensive even compared to other dissertations in the same field.
As I discuss a large amount of theory and design work, the length is somewhat justified, however, creating such a long work also put a strain on me as the author.
The work was almost exclusively written in 2025 between March and September with a pace of about 5 pages per day.
To all PhD students reading this, I can highly recommend to write a shorter dissertation.

My disseration can be found here: https://doi.org/10.53846/goediss-11674