•11 min
You Upgraded Your Embedding Model and Silently Broke Retrieval
Swapping the embedding model behind your RAG looks like a one-line config change. It is a full data migration with semantic consequences, and it fails without ever throwing an error: query vectors from the new model and document vectors from the old one live in different geometric spaces, so retrieval quietly returns the wrong chunks. Here is how to reindex with a versioned dual index, gate the cutover on a labeled retrieval eval, and keep an instant rollback.
RAG
Embeddings