CHAISE - AI engineering made practical with the Tool Chain for AI Systems Engineering

Challenging development of AI systems

Like all software, AI applications are subject to constant change. In hardly any case the area of application, the desired functionality and all framework conditions are fully known at the start of the project. Over time, especially when the system is in operational use, new requirements and customization requests arise. In the case of machine learning applications, additional challenges make a structured approach particularly important, both in the initial development phase and during operation, as well as in the further development of software components and the retraining of machine learning models. This is the only way to ensure that (further) developments can be transferred beyond the prototype stage into operational environments.

AI engineering as a new discipline is dedicated to these challenges. It addresses the systematic development and operation of AI-based solutions as part of systems that perform complex tasks. It enables the professionalization of AI system development and makes the process plannable, reliable, traceable and reproducible.

A necessary building block is a structured way of working. PAISE®, the Process Model for AI Systems Engineering, developed under the leadership of Fraunhofer IOSB, defines a suitable process model and forms a tried-and-tested methodological basis for AI engineering.  

CHAISE, a Toolbox for MLOPs

Another building block is the tools and instruments required for the development, training and evaluation of AI systems - in the structured approach mentioned above, in line with the MLOps paradigm (based on DevOps, which is familiar from software development). The aim of an internal R&D project was to implement MLOps in the best possible way based on our experience in many AI application domains and to automate it as far as possible. This has now resulted in CHAISE, the Tool Chain for AI Systems Engineering - an easy-to-use and expandable toolbox for creating machine learning applications and training ML models.

CHAISE, a Toolbox for MLOPs

The software suite integrates a unique set of tools for data management and AI model development, storage, and execution. This integration enables, among other things:

• The versioning of different model versions, including their quality criteria
• Support for the transition to operation and subsequent monitoring, for example through model drift detection (using the AutoDrift tool)
• (Sensor) data management in accordance with the OGC standard (using FROST®)
• Video data annotation (ANTONN)
• The explainability of AI procedures (XAI toolbox)

Routine activities such as the selection of tools and their time-consuming installation on server systems are greatly simplified and automated, leaving more time for core tasks such as method selection and model development.

CHAISE can be used on all systems on which a Kubernetes-based container environment is available. This means that CHAISE can be used on a developer system, on-premise in your own data center or in the cloud. Of course, CHAISE can be adapted to your own needs and use cases and is designed to be open so that other tools can be added without any problems.