Worked on stabilizing Kubeflow MNIST deployments in air-gapped environments within the HPEEzmeral/aie-tutorials repository, focusing on improving image resolution and deployment reliability. Addressed a critical bug by correcting Docker image tags and registry paths for the Katib and TFJob launchers, ensuring versioned and accurate image references. Refactored the airgap registry path to include the tutorial image prefix, which enhanced reproducibility and reliability for on-premises and offline setups. Leveraged Python, YAML, and Docker to deliver these improvements, resulting in smoother customer-facing tutorials and more dependable Kubernetes and Kubeflow workflows, particularly in restricted or offline network environments.