BioOps: Enabling MLOps for scalable and flexible model-centric process control in biopharmaceutical development

The deployment of advanced modeling and machine learning in biopharmaceutical process control has been limited by legacy fragmented automation infrastructure and a lack of model lifecycle governance. We present BioOps, a modular automation framework for biopharmaceutical advanced process control (APC). The framework incorporates established Machine Learning Operations (MLOps) principles, including version-controlled model registries, containerized deployment, and continuous integration/continuous delivery (CI/CD) pipelines. BioOps explicitly decouples model development from model execution within the control architecture. This decoupling enables process scientists to deploy, test, and iteratively refine mathematical models and control strategies in a flexible manner. These workflows can be applied across heterogeneous bioreactor platforms and are executed entirely in-house, without reliance on vendor-specific solutions or custom system integration. In addition, BioOps embeds traceability, auditability, and model lifecycle management directly into the control stack. As a result, the framework aligns with regulatory expectations while continuing to support agile, in-house experimentation and development. Case studies across a geographically distributed set of bioreactors demonstrate BioOps' versatility in PAT-driven feedback control, hybrid model-based nutrient and glucose feeding, adaptive phase transitions, and cross-scale deployment. By operationalizing model-centric control through an MLOps-inspired architecture, BioOps provides a practical foundation for scalable, reproducible, and future-ready biomanufacturing.