Building with AI

Embedding

Turning words or documents into a list of numbers so that similar meanings end up close together.

In everyday terms

"Puppy" and "dog" get nearby numbers; "puppy" and "invoice" are far apart. That lets software search by meaning, not just exact keywords.

For professionals

Dense vectors from a trained encoder where geometric distance (often cosine similarity) approximates semantic similarity.

Think of it like…

A map of meaning: every idea gets coordinates, and related ideas live in the same neighbourhood.

You've already seen it

Search that finds "cheap flights" when you typed "budget airfare".

Myth vs reality

Myth: Embeddings are a kind of encryption.

Reality: They represent meaning so it can be compared, not to hide it.

Quick check

In an embedding space, "king" will be closest to…

• "queen"
• "banana"
• "invoice"
• "blue"

Builds on

Token

Related

Vector Database · Retrieval-Augmented Generation (RAG) · Token