05 · Embeddings, RAG & Vector Search · Topic 03

What is Vector Search?

Vector search finds items whose vector representations are close to a query vector.

What is it?

Vector search finds stored vectors that are most similar to a query vector.

The vectors usually come from an embedding model. A search system compares them with a distance or similarity measure and returns matching records, often with metadata and source text.

Think of it like...

Think of it like finding nearby places on a map.

A query is placed on the same map as stored items. Nearby points are possible matches. The analogy explains proximity, not literal geographic distance.

How it works

1 · EncodeCreate a query vectorConvert the user's query using an embedding model.

2 · CompareMeasure distanceCompare the query with stored vectors.

3 · RankOrder candidatesPut the closest matches first.

4 · FilterApply constraintsUse metadata, permissions, or freshness rules.

5 · ReturnSend resultsReturn records for retrieval or generation.

Real-world examples

Everyday example · Help article

Input: “How can I change my password?”

AI / System: searches nearby article vectors

Output: the most relevant help articles

Business example · Product catalog

Input: a product description

AI / System: finds similar catalog vectors

Output: matching products for review

What it is NOT

Vector Search ≠ Keyword Search

Vector search: compares learned representations.

Keyword search: matches literal terms or text patterns.

Vector Search ≠ Embedding

Vector search: finds similar stored vectors.

Embedding: creates the vector representation.

Vector Search ≠ RAG

Vector search: returns similar records.

RAG: uses retrieval results as model context for generation.

Remember this

Vector search finds nearby vector representations; it does not generate the final answer by itself.

Independent explainer

Video

visual explainer