Question
You have an Azure SQL database that contains tables named dbo.ProductDocs and dbo.ProductDocsEmbeddings. dbo.ProductDocs contains product documentation and the following columns:
- DocID (int)
- Title (nvarchar(200))
- Body (nvarchar(max))
- LastModified (datetime2)
The documentation is edited throughout the day.
dbo.ProductDocsEmbeddings contains the following columns:
- DocID (int)
- ChunkOrder (int)
- ChunkText (nvarchar(max))
- Embedding (vector(1536))
The current embedding pipeline runs once per night.
You need to ensure that embeddings are updated every time the underlying documentation content changes. The solution must NOT require a nightly batch process.
What should you include in the solution?
- A. fixed-size chunking
- B. a smaller embedding model
- C. table triggers Correct
- D. change tracking on dbo.ProductDocs
Explanation
Microsoft recommends updating embeddings every time the underlying data changes. An AFTER INSERT, UPDATE trigger on dbo.ProductDocs fires with each change and can regenerate the affected chunks and embeddings (for example with AI_GENERATE_EMBEDDINGS) immediately, with no scheduled batch. Change tracking only records which rows changed; something still has to poll it and do the work, which is again a scheduled process. Chunking strategy and model size affect embedding quality and cost, not when embeddings are refreshed.
