Within the SAFE-6G architecture, the MLOps platform supports the preparation, training and deployment of Machine Learning models used by several key project components, including the SAFE-6G Chatbot, Cognitive Coordinator and AI-powered Trust Functions.

The platform provides an integrated environment covering the AI/ML lifecycle through several complementary capabilities: a Pipeline Development Environment for experimentation and model development, a Pipeline Orchestration Platform for automated ML workflows, Model Storage for datasets and trained models, Model Serving for real-time inference, and a Differential Privacy module for privacy-preserving Machine Learning.

Built around cloud-native technologies and Kubernetes, the SAFE-6G MLOps environment supports traceable model management, systematic versioning and automated deployment. This makes it possible for project partners to move AI models from experimentation to operational execution while maintaining consistency and reproducibility across different environments. One of the important developments described in D3.5 is the Model Serving component, designed within SAFE-6G to bridge the gap between Machine Learning development and operational deployment.

Trained AI/ML models can be packaged into Docker images and Helm charts and deployed automatically in Kubernetes environments. The platform uses technologies including KServe, GitLab CI/CD and Meta-OS integration to create a controlled pathway from trained model artefacts to production-ready inference services.

Model versioning also provides traceability, allowing SAFE-6G components to identify the exact version of a deployed model and, where required, redeploy previous versions in a reproducible manner.

This approach helps simplify the operational side of AI deployment, enabling developers to concentrate on model development while the platform supports the processes required to move models towards real-world use. 

Within the SAFE-6G architecture, the MLOps platform supports the preparation, training and deployment of Machine Learning models used by several key project components, including the SAFE-6G Chatbot, Cognitive Coordinator and AI-powered Trust Functions.

The platform provides an integrated environment covering the AI/ML lifecycle through several complementary capabilities: a Pipeline Development Environment for experimentation and model development, a Pipeline Orchestration Platform for automated ML workflows, Model Storage for datasets and trained models, Model Serving for real-time inference, and a Differential Privacy module for privacy-preserving Machine Learning.

Built around cloud-native technologies and Kubernetes, the SAFE-6G MLOps environment supports traceable model management, systematic versioning and automated deployment. This makes it possible for project partners to move AI models from experimentation to operational execution while maintaining consistency and reproducibility across different environments. One of the important developments described in D3.5 is the Model Serving component, designed within SAFE-6G to bridge the gap between Machine Learning development and operational deployment. Trained AI/ML models can be packaged into Docker images and Helm charts and deployed automatically in Kubernetes environments. The platform uses technologies including KServe, GitLab CI/CD and Meta-OS integration to create a controlled pathway from trained model artefacts to production-ready inference services.

Model versioning also provides traceability, allowing SAFE-6G components to identify the exact version of a deployed model and, where required, redeploy previous versions in a reproducible manner. This approach helps simplify the operational side of AI deployment, enabling developers to concentrate on model development while the platform supports the processes required to move models towards real-world use.

Find it online: https://safe-6g.eu/wp-content/uploads/2026/09/D3.5_SAFE-6G_V1.0.pdf