Lesson 3.1:
Vector Search & RAG

Semantic Search with Vectors

Traditional keyword search is being replaced by semantic vector search. By converting text into high-dimensional vectors (embeddings), we can find records based on *meaning* rather than exact matches.

Atlas Vector Search Index

First, you must define a vector index in your Atlas cluster. This allows MongoDB to perform lightning-fast nearest-neighbor searches.

{
  "fields": [
    {
      "type": "vector",
      "path": "embedding",
      "numDimensions": 1536,
      "similarity": "cosine"
    }
  ]
}

Retrieval-Augmented Generation (RAG)

RAG pipelines fetch relevant documents from your database and feed them to an LLM to provide context-aware, accurate answers. MongoDB Atlas serves as the "Long-Term Memory" for your AI agents.

// Example RAG flow
const context = await collection.aggregate([
  { $vectorSearch: { ... } }
]).toArray();

const response = await llm.complete({
  prompt: `Use this context: ${JSON.stringify(context)}. Question: ${userQuery}`
});

Check Your Knowledge

Which similarity metric is most commonly used for text embeddings?

• Cosine
• Euclidean