How to deploy a TensorFlow model for access through an HTTP endpoint?

I’ve done a few using built-in AWS algos but I’m unclear on how to package and deploy a custom TF model for access through an HTTP endpoint.

For instance, I'm using a semantic clustering algo and I will want to a) test semantic similarity between two inputs and b) do a semantic search for N entries that are similar to one input.

In a Juptyer notebook I can do all this wonderfully. But I need to be able to access from HTTP or from an AWS SDK.

Any help would be greatly appreciated!