Your production RAG system has 50M embeddings generated with text-embedding-ada-002. You want to migrate to text-embedding-3-large for better retrieval quality. You cannot take downtime and cannot afford to have mismatched embeddings (old and new models in the same index). Design the migration strategy. Cover: how to run old and new embedding models in parallel, how to handle writes during migration, index versioning and blue-green deployment for vector stores, how to validate retrieval quality before full cutover, and how long this migration realistically takes.