05 · Embeddings, RAG & Vector Search · Topic 05

What is Retrieval?

Retrieval selects relevant information from a larger source collection for a task.

What is it?

Retrieval is the process of finding and selecting information that is relevant to a query.

A retrieval system can use keywords, vectors, metadata, filters, or a combination of methods. It returns evidence for a person, an application, or a model.

Think of it like...

Think of it like finding the right pages in a library.

The library holds many pages, but a question needs only a small selection. Retrieval chooses that selection before the next step uses it.

How it works

1 · QueryRead the requestIdentify the information needed.

2 · SearchFind candidatesUse text, vector, metadata, or hybrid search.

3 · RankOrder resultsPut the most useful candidates first.

4 · FilterApply constraintsRespect access, freshness, and source rules.

5 · ReturnSend evidencePass selected records to the next step.

Real-world examples

Everyday example · Help center

Input: a question about an account

System: retrieves relevant support articles

Output: a short list of useful sources

Business example · Policy review

Input: a policy question and user permissions

System: retrieves allowed, current documents

Output: evidence for an answer or review

What it is NOT

Retrieval ≠ Search Results Only

Retrieval: selects information for a downstream task.

Search results: may be shown directly without further processing.

Retrieval ≠ Generation

Retrieval: finds existing information.

Generation: creates new model output.

Retrieval ≠ RAG

Retrieval: is the selection step.

RAG: combines retrieval with model generation.

Remember this

Retrieval selects useful existing information before another system or model uses it.

Independent explainer

Video

visual explainer