We've been doing some cool machine learning work with our risk model at @River. It's running directly in the Erlang VM using ONNX and Native Implemented Functions. This allows us to avoid the complexity of a distributed system. We should write a blog post about it.