Embedding

A numerical fingerprint of a piece of text that captures what it means rather than which words it used. Texts about the same thing get similar fingerprints even with no words in common — "MRI results" and "scan findings" land right next to each other.

Filing by subject rather than by spelling. Two people describe the same fault with no shared vocabulary; sort by meaning and both notes end up in the same drawer.

Why it matters: this is the machinery that lets AI search by meaning instead of exact keywords. It's why a RAG system (the open-book-exam recipe, where the AI answers from your documents) still finds the right passage when the person asking phrased it completely differently from the way it was written down.

Related: Semantic search · Vector / vector database