Rustling up predictive sporting betting models on the BEAM

Rustler is a library that makes easy to bind Rust code to the BEAM as a NIF. At SimpleBet, David and his team took that to the next level by implementing our machine learning models in Rust as an application-level NIF. Affectionately referred to as the Dream Stack, they've used this approach to build a system that generates the odds of every plate appearance outcome at a baseball game. While it worked well to start, they are now migrating to a service-based approach instead.

THIS TALK IN THREE WORDS

Very

Big

NIF

OBJECTIVES

• Teach the audience how/why to write NIFs using Rustler
• Demonstrate SimpleBet's approach to using NIFs and how we took it too far
• Explain why that approach will no longer work well for SimpleBet and why moving to a service-based approach, while slower, aligns much better with our company goals

TARGET AUDIENCE

The audience for this talk are those who are interested in Rust and machine-learning, but most importantly, those who want to learn from about an architecture that worked really well until it didn't, and how to migrate away from that type of situation.