Build an AI copilot with vCore-based Azure Cosmos DB for MongoDB and Azure OpenAI: Review

The first three modules of this path are just a user manual. Click here, copy this, paste that. It’s all about provisioning the vCore-based Cosmos DB, migrating data, and managing the cluster. Zero thinking required. I could have gotten the same information faster from the official docs. The whole time I was wondering when the actual 'AI copilot' part would start.

It finally shows up in the last module. And it’s a pretty standard, minimal RAG setup. You get a Jupyter notebook, a sample dataset, and some Python code that wires up Azure OpenAI to the Cosmos DB vector store. The code works. You run it, and it answers questions about the data. Done. Copilot built.

But it doesn't teach you anything. The code is just handed to you. There's no explanation of why you'd choose a specific chunking strategy, what a better embedding model might do, or how this would fail with a more complex dataset. It's a happy-path demo. All the hard parts of building a real RAG system—the parts that actually matter in production—are completely ignored.

So, what is this? It's a sales demo for the Cosmos DB vCore SKU, showing that it can, in fact, be used as a vector database for a basic AI app. If you're an Azure dev who just needs to see the wiring diagram for these specific services, it saves you an afternoon of fumbling. If you actually want to learn how to build AI applications, look elsewhere.