Uploaded August 2026 | Updated September 2026, 1 week ago

---

Application data changes, but its vector embeddings do not update by themselves. Stale embeddings can reduce semantic-search relevance and hurt the quality of results returned by RAG applications.

Some techniques include:

- Generate embeddings in the application's write path
- Regenerate them asynchronously using a change feed and background worker
- Let the data platform manage updates (for instance, with Integrated Embeddings in Azure Cosmos DB)

Learn more: https://aka.ms/embeddings-cosmosdb

00:00 Why embeddings need to stay synchronized
00:27 Option 1: Generate embeddings on write
00:45 Option 2: Update embeddings asynchronously
01:29 Option 3: Managed updates with Azure Cosmos DB
02:28 Choosing the right approach

#RAG #VectorSearch #AzureCosmosDB