Why should you care?

“Find lessons like this paragraph” needs meaning, not only keywords. AI features — recommendations, RAG chat — depend on vector search.

See it live — copy this example

Open mongosh or MongoDB Compass, select database nosqlverse, then run the example. Change one field and run again.

db.lessons.insertOne({
  title: "Replica Sets",
  text: "High availability with primaries and secondaries",
  embedding: [0.12, -0.44, 0.09] // short demo vector; real ones are 768–1536 dims
})
// Atlas Vector Search index (JSON config in UI) on embedding
// Example aggregate stage shape:
db.lessons.aggregate([
  {
    $vectorSearch: {
      index: "lesson_vector_index",
      path: "embedding",
      queryVector: [0.11, -0.40, 0.10],
      numCandidates: 50,
      limit: 5
    }
  },
  { $project: { title: 1, score: { $meta: "vectorSearchScore" } } }
])

Run Example »

Edit the code below and click Run to see the result in Toolliyo’s live editor.

What happened?

• Documents store embeddings from an ML model.
• $vectorSearch finds nearest vectors using an Atlas vector index.
• Scores rank semantic closeness.
• Keep dims consistent with the model.

Practice next

1. Create a small lessons collection with fake embeddings.
2. Define an Atlas Vector Search index on embedding.
3. Run a $vectorSearch aggregate.
4. Add a $match pre-filter on courseId if supported by your index filters.
5. Compare keyword search vs vector results on the same prompt.

Remember

Vectors capture semantic meaning. Atlas indexes them for ANN search. Use $vectorSearch in aggregations.

RAG tutor for NoSQLVerse

Learner asks a question; backend embeds it and retrieves top lessons as context for an LLM.

Outcome: Answers cite the right MongoDB topics instead of hallucinating APIs.