Lesson 2 of 7
Embeddings
6 min read
"Cancel my plan" and "how do I end my subscription" share almost no words. To a keyword search they're strangers. So how does retrieval know they mean the same thing?
Meaning as a point in space
An embedding turns a piece of text into a list of numbers — a vector — that acts like coordinates for its meaning. Text with similar meaning lands in nearby spots; unrelated text lands far apart. The model learned this map by reading enormous amounts of text.
An embedding places text in a space where distance means similarity — close points mean similar things.
Search by meaning, not words
Once every passage is a point, finding relevant text is just finding nearby points. "Cancel my plan" lands right next to "end my subscription" even with no shared words — so retrieval matches on meaning. This is why RAG can answer a question phrased nothing like the document that answers it.
Nearest points = closest in meaning. That's how retrieval finds the right passage even when the words don't match.
Real embeddings have hundreds or thousands of dimensions, not two — but the intuition holds exactly: closer means more similar. The 2D picture is just that map flattened so you can see it.
The shape of it
- —An embedding is a vector — coordinates for a piece of text's meaning.
- —Similar meaning lands in nearby points; unrelated text lands far apart.
- —Searching by nearest points matches meaning, not exact words.
A user asks "how do I get my money back?" Your docs only ever say "refunds". Why does embedding-based retrieval still find it?
Continue in the app
Take the whole RAG & Search course — tracked
Get your personalized path, progress and streaks in the app — this lesson and every next one, in order.