Tutorials · Chapter C (3/4) · ~7 min
Vectors & similarity search
Play → read → next
Similar meanings have vectors near each other — that’s how AI finds related text.
Playground
Vectors & similarity
Compare two phrases, then pick the best matching document.
Similarity 77
Rank docs for: How do I reset my password?
Recap
What you just did
You ranked documents by similarity to a query — not by exact keyword match. The closest meaning wins.
Teach
How it works
- Embed the question
- Embed (or look up) each candidate chunk
- Score closeness (often cosine similarity: do the arrows point the same way?)
- Return the top matches
That’s “search by meaning.” Keywords can still help, but meaning covers synonyms and paraphrases.
Use it
When you'd use this
- FAQ bots that find the right article even wording differs
- “People also asked” style related content
- Building blocks of RAG (next lessons)
Watch out
Watch out
Highest similarity can still be the wrong doc if your collection is noisy. Retrieval quality beats model cleverness.
Try next
Try this next
Write a question using different words than the answer article. Good similarity search should still find it.