You will build a model with BigQuery ML and host it on Vertex AI. Vendor streams arrive continuously and may contain invalid values. For near-real-time processing and sanitization before BigQuery, what should you do?

The correct answer is to use Pub/Sub with Dataflow. Pub/Sub provides a scalable, asynchronous messaging service for ingesting continuous streams of data from multiple vendors. Dataflow is a fully managed service for executing Apache Beam pipelines, which is ideal for near-real-time processing, sanitization, and transformation of streaming data before it's loaded into BigQuery. This approach ensures data quality and prepares it for BigQuery ML. The other options are less suitable: Using BigQuery streaming inserts directly into a dataset for an ML model doesn't allow for sanitization before ingestion. A Cloud Function might be suitable for simpler, lower-volume transformations, but Dataflow is designed for high-volume, continuous streaming data processing and offers more robust error handling and scalability for complex sanitization logic.