The Invisible Scaffolding of MLOps (and my new Ebook)

“Success is not the model you deploy—it’s the system that keeps deploying it.”

I started this project as a curiosity: Could I actually build a system that lives on all three major clouds without losing my mind?

The answer is yes, but the path was filled with broken pipelines, IAM permission errors, and a lot of debugging.

View to GitHub Repo

I realized that Identity is the new perimeter. When you move from one cloud to three, you can’t rely on firewalls alone. You need federated trust.

I have compiled my entire journey—the architecture diagrams, the Terraform patterns, and the observability strategies—into a new digital release: “Building Resilient ML Pipelines Across GCP, AWS & Azure.”

What’s inside:

• Chapter 2: Why I ditched static keys for Workload Identity Federation.

• Chapter 6: How I normalized logs from JSON (AWS) and Proto Payloads (GCP) to speak the same language.

• Chapter 7: The actual cost breakdown (spoiler: Serverless kept it cheap).

View the full PDF below:

View the Documentation