Consider adding S3/GCS/etc buckets as configurable sources and sinks for requests and responses.
Possible features:
A batch API: An API for kicking off a batch job that runs across all files in a bucket path. API could allow users to kick off runs with different models against in-place data.
Watching a bucket for requests, optionally removing the objects after they are processed. (Might require polling).
Using a bucket as the location where responses could be written to when requests are consumed from a messaging topic/queue.
Consider adding S3/GCS/etc buckets as configurable sources and sinks for requests and responses.
Possible features: