PRODUCTION SCENARIO
A Generative AI Engineer indexes a policy table that the compliance team edits a few times a week. The assistant must reflect an edit within the hour, and the team does not want to pay for compute that sits idle between edits.
How should the engineer configure the index?
Answering here is anonymous. Nothing is saved unless you sign in.
Show answer and explanation
Answer: Triggered sync mode, started by the job that publishes each edit
With triggered sync mode you use the Python SDK or the REST API to start the sync, so compute runs when a sync is requested, and starting it from the publishing job lands each edit inside the hour. Continuous sync holds the index within seconds of the table but provisions a cluster for the streaming pipeline, which is the idle cost being avoided. A nightly job is the right mode on the wrong trigger and misses the hour budget by most of a day. Direct Vector Access hands the application embedding and upsert work Delta Sync already does.
Official Source: https://docs.databricks.com/aws/en/ai-search/create-ai-search ↗