High Throughput ML Pipelines and Predictions in Production Systems

Ravelin is a smart fraud detection and prevention platform that helps companies stop online payment fraud by examining customer behaviour data and spotting fraudsters while there is still time to block them. The company imports a client’s visitor, registration, and payment data in real time, via an API, inspects data using an AI, identifies and blocks fraudsters, and enables systems to prevent such crimes in future.

Ravelin runs a go microservices platform that makes around 1,000 predictions a second. This talk will give a walk through our multiple attempts to scale and optimise model deployment. A journey that will take use through Docker, CoreOS Rkt, Pickel, Protobuf, Sidecars, Python and Go.

One thought on “High Throughput ML Pipelines and Predictions in Production Systems

Leave a Reply