Batch processing assumes "we'll process accumulated data later all at once." Streaming processing assumes "we process data the moment it arrives." The difference isn't just fast vs slow. Batch assumes data has boundaries (files, date partitions). Streams handle infinite unbounded data. To compute something like "today's revenue total" from endless data, you have to draw a boundary somewhere artificially. That boundary is called a window.
In AWS, Amazon Managed Service for Apache Flink handles this streaming aggregation. Formerly "Kinesis Data Analytics," it had SQL and Flink variants; the SQL version (KDA for SQL) stopped accepting new creations in 2023 and Apache Flink is now standard