Data capture on sagemaker severless inference endpoints

Hi,

Here are my thoughts, I haven't yet validate this.

If you want to do data capture on a serverless inference endpoint you would need to build own solution that mimics what the data capture feature does namely capturing the API requests, converting it to the same format as the data capture feature and storing it in S3. You can get this format from deploying a standard endpoint and monitoring its output.

This would mean that your inference container will need to to serialise and write each request to S3 in the same format as standard monitoring job would.

After the data is in S3 you would need to manually kick off the data/model monitoring jobs.

An alternative (non-AWS way) would be to look at directly integrating a 3rd party service within your Inference container such as whylabs.ai