-should u not have any data to work on, then answer a question for which there is some public data

-models doing unexpected outcomes while training on the job

-Glue :: spark pseudo cluster sevice triggered on schedules

-athena :: imposes schema on the files and does SQL, managed presto

-EMR

-Spectrum ::   join data in S3 from within the warehouse along with data in the enterprise data warehouse that is Redshift, inhouse offering of aws

-orc parque format

-Sagemaker a more Notebook instances

-EDA ready

?? Provision container based on load??

auto scaling, model latency,

You dont get to touch the containers

AB testing doable::: multiple models sitting behind the endpoint