Problem
A fraud model relies on a feature; "transactions in the last 60 seconds"; that was being computed by the same nightly batch ETL job that handles everything else. By the time it updated, fraud losses had already spiked 40% for the day. Give this feature a real-time path that the nightly batch job can never touch, fast enough to serve roughly 5,000 fraud-score requests a second, and make sure a lagging real-time path degrades to a stale-but-available value instead of blocking a prediction outright.
Required components
Click to add a component to the canvas.
Additional components
Browse by category. Click or drag a component onto the canvas.