I have started to dig into AWS Personalize for a project. In reality I have been dying to get hands on with it for a while but only recently has it become viable.

AWS Personalize is at its core a recommendation engine. It trains a model that will predict which products the user might like based on user behaviors. The easiest example to grasp is their video recommendation service.

If it observes that you watch a lot of Steven Seagal movies it will learn that you have bad taste in movies and recommend accordingly (JK).

Here are some of my finding:

Terraform: I was unable to find a terraform resource for this in the main AWS terraform module but managed to find one in the AWSCC module.

Pricing: At $0.15 per 1000 inferences it is not cheap.

For e-commerce companies with a lot of spam bots I recommend limiting giving recommendations to only users that are logged in and email verified. Or, better yet, have actually bought something from you in the past. You will have to decide for your specific use case.

Kinesis Usecase: Interesting enough I have not found a way to directly pipe events from Kinesis into Personalize but I did find usecases where you simply have a Lambda consume the Kinesis event and pipe it into Personalize. I am sure they will launch an integration for that the second I finish writing the code to do it programmatically.

Custom Usecases: Evidently the main 2 use cases are the product recommendation and media recommendation, but there is a third option to train custom models. This I find really interesting and want to dig in on more.

Question: What would you use AWS Personalize for? Any creative use cases out there?