A podcast platform records high-volume user events and runs an anomaly detection model using a 10-minute running window. Each event needs small transformations before inference. Which event ingestion and pre-processing design provides the required transformations with the least operational overhead?

The correct option leverages Kinesis Data Streams for high-volume event ingestion and Kinesis Data Analytics (Apache Flink) for real-time transformations and windowing. Flink is ideal for continuous data processing, including windowing operations like the 10-minute running window required, and offers managed service benefits, minimizing operational overhead. Incorrect options: AppSync, DynamoDB, and Lambda introduce unnecessary complexity and latency for high-volume, real-time stream processing. DynamoDB Streams are not designed for continuous, windowed transformations across a stream. Kinesis Data Firehose to S3 is suitable for batch processing or archiving, not real-time, low-latency transformations needed for a running window. AWS Glue is a batch ETL service, not optimized for real-time stream processing. Amazon MSK is a powerful streaming platform, but using a Lambda function for continuous, windowed transformations on a high-volume stream would be less efficient and incur higher operational overhead than a dedicated stream processing engine like Flink.