Towards a structured AI development lifecycle for reusable AI products in the public sector

Currently, AI-enabled solutions in the public sector are integrated in a fragmented way that does not allow them to be easily transferable to other public organisations, highlighting the need for methodologies that streamline AI development process. For this reason, regulatory bodies have established standards that define what characteristics AI systems need to meet. Meanwhile, recent research initiatives call for comprehensive methods that offer guidance on howtoembedrequirement aspects into the main AI development stages. However, existing approaches offer high-level recommendations overlooking the difficulties in translating them into lowlevel implementations in order to bring together procedural and technical approaches. Moreover, the level of granularity of AI lifecycle stages does not facilitate the modularity of AI systems, which would enable the easy integration of new implementation to meet system requirements. In this paper we enhance the AI lifecycle pipeline with explicit operations that aid in the modular development and reuse of AI systems. Furthermore, we propose a structured framework that enables the operationalisation of system-wide objectives throughout the various phases of the AI lifecycle, while systematically leveraging existing toolkits. To illustrate the framework’s adoption, we present a proof-of-concept example to showcase its practical application.