How would you architect a data pipeline to ensure real-time model inference for a high-traffic client at Braze?
