DevOps for AI: More than Just Models – Success through Collaboration and Automation
Abstract
In today’s enterprise landscape, MLOps (DevOps for AI and ML) and collaboration are crucial for moving AI from proof-of-concept to production. This talk focuses on the process and collaboration aspects of implementing sustainable AI. We’ll explore how to establish effective stakeholder engagement and showcase how the MLOps process can be implemented and enhanced through AI automation, including the use of AI judges for model evaluation.
The presentation highlights practical implementation strategies using vendor-neutral tools such as CNCF’s Kubeflow, which are effective across various environments. This is essential for enterprises operating in multi-cloud settings. Special emphasis will be put on the human-AI collaboration workflow and how AI can streamline operational processes.
This talk was held in German at IT-Tage 2025 on December 11, 2025 in Frankfurt am Main, Germany.
MLOps - Machine Learning Operations
Key Takeaways
- Don’t throw out what you learned from classical IT projects: requirement engineering, phases, KPIs
- At the end people need to talk! DevOps + MLOps + Any-Ops
- Always evaluate your results—outcome can change dramatically after small changes
- Start small: Kubeflow has a lot of components, start with one and avoid over-engineering
- It is not a magic platform—you still need to think about the process
Links
Related Resources
- Kubeflow - The Machine Learning Toolkit for Kubernetes
