PI: Grzegorz J. Nalepa, School of Information Technology, Halmstad University
co-PI: Sule Tekkesinoglu, Department of Computer Science, Lund University
State-of-the-art software engineering (SE) methods fail to adequately support AI-based systems that must operate reliably under dynamic conditions, where multiple heterogeneous components produce evolving and often conflicting outputs. While Machine Learning Operations (MLOps) attempts to bridge this gap, it lacks operational frameworks for transparent, continuous governance of multi-model AI systems, particularly in safety-critical contexts. This PhD project proposes AXLOP, an Agent-based Explainable MLOps environment, designed to address these challenges through coordinated agent architectures that generate, evaluate, and adapt alternative pipeline configurations, handle interactions between heterogeneous components and conflicting outputs, and make them explainable throughout the MLOps lifecycle. AXLOP will be developed and validated in collaboration with industry partners in the context of safety-critical domains, where regulatory constraints and safety requirements make robust AI governance essential.
Project humber: H6