← Back to glossary

RAG and search

Embeddings

Numerical representations that place text or other data with similar meaning close together in a multidimensional space.

Business example

Polp can find a leave policy even when an employee uses different wording from the original document.

How it works

1.A model converts content into vectors.
2.Vectors are stored with their source metadata.
3.A query vector is compared with stored vectors.

Common misconceptions

Embeddings contain a readable copy of the document.

They encode relationships for comparison; the original text and access controls remain separate.

Related concepts

Vector database Semantic search Retrieval-Augmented Generation (RAG)

Related reading

RAG is not uploading PDFs: permissions, freshness, traceability, and reliable sources

Sources

https://platform.openai.com/docs/guides/embeddings

Reviewed: 2026-08-04

Reviewed by: Javier Chulvi Bernad · LLM Engineer · Madrid