The capabilities of pre-trained Large Language Models (LLMs) open up new possibilities for the development of recommendation systems. However, data protection requirements, limited data availability, and last but not least the black-box nature of LLMs present hurdles for their use.

We show how Explainable AI and feedback mechanisms support self-determined use of the recommendation system. In addition, we present strategies for data protection-compliant and cost-effective operation of LLM-based solutions.

A structured approach for the development of explanation components has proven successful in numerous projects in research and practice.

In four consecutive phases, we first capture the target group and application context, then identify suitable XAI methods, which we then test with a prototype and finally develop into an application-ready component.

Modern AI systems have the reputation of being black boxes, the functioning of which remains hidden from users and developers. On the way to future use of AI, companies are therefore faced with challenges that have not previously arisen in classical software development.

In an article for heise online, our co-founder Kilian Kluge and INNOQ expert Isabel Bär show how companies can build their AI software on the three pillars of MLOps, Model Governance, and XAI to effectively control and operate it in accordance with legal requirements.