In this age of rapidly advancing progress in the development and range of applications of machine learning (ML) models, as well as an ever expanding catalog of frameworks and tooling [Huy20], it is important to not disregard the interoperability of these tools. An observable industry trend is pushing for more automation in the ML lifecycle and flexible and reliable MLOps, with MLflow [ZCD+18] being one solution in widespread use today. Offering unified tools for the exploratory development of models and transparent tracking of experiments and model versions, MLflow’s extensive API also makes it suitable for integration with other tools and platforms. In his bachelor’s thesis “Development of a Deployment Platform for ONNX Models”, H. Megahed introduced the NEXON platform for reliable deployments of ML models in the form of manually uploaded ONNX files. Following the structured process of the Design Science Paradigm proposed by Runeson et al. [RES20], my work extends this platform through an integration with the MLflow Model Registry. The goal, to enable automatic deployments of models in the registry on NEXON, is realized through the approach of defining a desired deployment state and pulling the fixed model versions accordingly. The prototypical integration produced in this thesis introduces a synchronization API that downloads and deploys ONNX models from MLflow based on a list of semantic version selectors covering the version-control, tagging and aliasing features provided by the registry. Additionally, the “latest” option ensures deployment of the most recent versions. Through manual testing of functional requirements, my results confirm the applied approach as a valid technological rule for integrations of this kind. Finally, limitations of the implementation are addressed and possible future enhancements are proposed.