Engineering Data Architectures for AI/ML Integration in Regulated Manufacturing

Abstract

<p>Life science, i.e. pharmaceutical and medical device, manufacturers are increasingly exploring artificial intelligence (AI) and Machine Learning (ML) to enhance production quality and regulatory compliance. However, current data handling practices result in data fragmentation, and complex regulatory requirements present barriers to wide implementation. In this study, we conducted 20 qualitative interviews with data architects, AI specialists, and regulatory compliance officers. Our aim was to get a better understanding of the current state of the field, challenges and future outlook in regulated manufacturing, employing the Gioia methodology. Our findings highlight data silos and legacy infrastructures as primary technical barriers, while evolving regulatory frameworks and uncertainties in AI validation create significant compliance challenges. Interviewees emphasized the necessity of unified data architectures and platforms, embedded governance mechanisms, enhanced security, and proactive regulatory operations (RegOps) to enable both innovation and compliance. Based on the interview results, we propose a conceptual framework to guide the design of AI-driven data architectures that bridge fragmented systems and support compliant AI/ML lifecycle management. This study is the first phase of research efforts aiming to implement and validate AI/ML solutions grounded in industry needs.<br></p>