A lab's IoT sensors emit 100 KB payloads every 10 seconds. A downstream process polls an Amazon S3 bucket every 30 seconds to consume the data. Which architecture will deliver the data into S3 with the LEAST latency?

The correct option minimizes latency by using the Kinesis Client Library (KCL) with a 5-second buffer. KCL allows direct consumption from Kinesis Data Streams and custom buffering, ensuring data is written to S3 quickly, aligning with the 10-second sensor interval and 30-second polling. Using Kinesis Data Firehose with its default buffer (typically 60 seconds or 5MB) would introduce higher latency than necessary, as data would sit in the buffer longer. Configuring Kinesis Data Streams with 5 shards alone doesn't address the delivery mechanism to S3 or the buffering strategy, so it doesn't guarantee low latency delivery to S3. Amazon Managed Service for Apache Flink (Kinesis Data Analytics) is for real-time processing, not primarily for direct, low-latency delivery to S3, and adding Firehose with a 5-second buffer still introduces an additional service layer compared to KCL directly writing to S3.