What is an embedding?

In practice

Embeddings are what make semantic search work, and they are the first half of most RAG systems. They are also a silent source of drift: change the embedding model and every stored vector has to be regenerated, because old and new vectors are not comparable. Plan for that before you index a million documents.

Where it fits in a build

Use embeddings for meaning-based retrieval, deduplication, clustering support tickets, or matching an enquiry to the right service. Keep keyword search alongside them — exact terms, part numbers and reference codes are exactly what semantic search is worst at, and the combination beats either alone.

Common mistakes

• Relying on semantic search alone, then failing to find a document by its own reference number.
• Embedding chunks that are too large, so one vector represents four unrelated ideas and matches none of them well.
• No plan for re-indexing when the embedding model is upgraded.

Related terms

• Vector database
• Retrieval-augmented generation (RAG)
• Context window

Working on something that involves this?

ASTACKRA designs and builds AI systems, automation and custom software for businesses that need technology shaped around their own workflow. If this term turned up in a proposal and you want a straight answer about whether it applies to your situation, ask us — no obligation, and we will tell you if the answer is no.