An embedding turns text into a vector: a list of numbers that represents its meaning in a form a computer can compare. An embedding model creates these vectors for documents and questions. section 16.1.2
Similar meaning, different words
“Dog” and “puppy” use different words but have similar meanings, so their vectors would be similar. That lets a search find relevant text even when its words do not match the question. section 16.1.2
In a RAG application, embeddings help find documents to use when answering a question:
- Convert each stored document into a vector. section 16.1.2
- Convert the user's question into a vector. section 16.1.2
- Find documents with the closest vectors. This is called similarity search. section 16.1.2
Embeddings handle the comparison step of retrieval; they are not the answer themselves. They connect document search to the wider RAG pipeline. section 16.1.2 section 18.1
