Today the adoption of Machine Learning is enormous. We use it almost everywhere: on clusters, on our phones, on hand watches and even on the teapots. But, who knows how challenging could be the development of the continuously improved (or even real-time) machine learning. Servers’ capacity, stability, performance, along with a proper ML algorithm selection and many other properties must be taken into account. In this presentation, we will be demonstrating bots education for multiplayer Pacman game. We will show how elegant could be the development of such a problem with Kafka, Akka, RSocket and Reinforcement Machine Learning.