Planning

☐ Review this handy Vector Search overview with your team to get familiar with the basics.

See how other companies have successfully built AI apps on MongoDB with our AI Solutions Library.

☐ Check out the Vector Search Toolkit - a one-stop-shop for the most helpful Vector Search onboarding content. 

☐ Define your use case. Who are your users and what information are you trying to help them find?

Think about the different content owners for each document and consider how you'll ensure that information is kept up to date.

☐ Identify your document set. What content are you including in your application and where is it saved today?


☐ Start aligning on chunking strategy, embedding provider, and data modeling with your account team.


☐ Develop a test dataset that maps expected user queries to the right documents.

As you iterate on different retrieval strategies, this dataset will help you evaluate the accuracy of your application.

☐ Read up on LLM/AI integration points and review yours with MongoDB’s GenAI team (and/or partners).

Setting Up

☐ Review your deployment options and add Dedicated Search Nodes to your Atlas Cluster.

Do note that Vector Search can often utilize low CPU Nodes.

☐ Create your Vector Embeddings.

Get a full rundown on how to do so with this guide.

☐ Follow this how-to article to create your Vector Search Index.

Testing

☐ Get familiar with running Queries against your Vector Search Index.

☐ Use Exact Nearest Neighbor Search (ENN) during testing to guarantee that the closest vectors are always being returned.

When you feel confident in your embedding model/chunking strategy, you can also use it to benchmark accuracy of ANN queries.

☐ Troubleshoot any errors.

Learn how to troubleshoot with this article. You can also reach out to your MongoDB team for help!

Performance Tuning

☐ Become an expert at optimizing Vector Query Performance.

Again, your MongoDB team is happy to help you upskill here. Don’t hesitate to reach out!