Retrieval & grounding
What Are Embeddings?
The quiet technology underneath RAG, semantic search, and recommendation systems.
An embedding is a way of representing a piece of text, image, or other content as a list of numbers (a vector) that captures its meaning. Content with similar meaning ends up with similar numbers, even if the exact words are completely different.
This simple idea — meaning as geometry — is what makes semantic search, RAG, and many recommendation systems possible.
Key takeaways
Meaning as geometry
Imagine every sentence placed as a point in space, positioned so that sentences with similar meaning land near each other and unrelated sentences land far apart. That is what an embedding model does, just in a space with hundreds or thousands of dimensions instead of two or three.
Once meaning is represented this way, finding "similar" content becomes a mathematical problem: measure the distance between points.
Why this matters for search and retrieval
Traditional keyword search only finds exact or near-exact word matches. Embedding-based search finds content with related meaning even when the wording is completely different — asking about "cancelling a subscription" can correctly retrieve a document about "ending recurring billing."
This is the core mechanism that makes RAG and modern semantic search meaningfully better than older keyword-only systems.
Choosing and using an embedding model
Embedding models vary in quality, dimension size, cost, and the languages or content types they handle well. For most teams, using a well-regarded general-purpose embedding model from a major provider is a safe, effective starting point.
What matters more than the specific model is consistency: use the same embedding model for both your stored documents and incoming queries, or similarity comparisons will not make sense.
Put this into practice