Machine Learning Engineer @ Staffbit in Staffbit

Creation of a system that allows for the simulation of gameplay in many game economy configurations, where a simulated player - an agent - with autonomous decision-making skills interacts with the environment.

Using reinforcement learning artificial intelligence algorithms, in particular PPO and IMPALA. The trained AI agent will be a neural network that selects an action from a finite set that maximizes the expected reward. The agent will have limitations identical to the physical limitations of a real player, such as limited visible information and limitations on performing actions. The reward will be calculated based on your game score. The trained agent will allow you to evaluate the given configuration of the game economy in which it was trained.

An optimal configuration will be selected in terms of the rules of good game design, which will be defined in consultation with game design experts, so that the player reaches the expected level in the game after a certain time or a certain number of interactions.

The goal of the next phase is to obtain a model that will improve monetization by at least 10%. The end result of the project will be the implementation of innovative technology developed in the course of R&D work in the form of a set of tools for simulating the game economy without the need to use historical player data, to support the game designer in selecting optimal game parameters and to generate personalized offers. The developed technology will combine techniques for storing and analyzing large data sets and machine learning.

• Development of an AI simulator structure that will be able to simulate various scenarios and configurations based on input data
• Selection and implementation of algorithms as an optimization method
• Implementation mechanisms that reduce the number of configurations necessary to test
• Analysis of model performance in terms of speed
• Creation of machine learning models to simulate player behavior
• Work related to input data to the model about paying users and their categorization
• Generating artificial input data for the second task using AI models
• Gameplay simulations using artificial intelligence
• Work related to creating the system personalized offers
• Validation of models based on historical data

Creation of a system that allows simulation of gameplay in many game economy configurations, where a simulated player - an agent - with autonomous decision-making skills interacts with the environment.

Using reinforcement learning artificial intelligence algorithms, in particular PPO and IMPALA. The trained AI agent will be a neural network that selects an action from a finite set that maximizes the expected reward. The agent will have limitations identical to the physical limitations of a real player, such as limited visible information and limitations on performing actions. The reward will be calculated based on your game score. A trained agent will allow you to evaluate a given configuration of the game economy in which he wasrespected.

An optimal configuration will be selected in terms of the rules of good game design, which will be defined in consultation with game design experts, so that the player reaches the expected level in the game after a certain time or a certain number of interactions.

The goal of the next phase is to obtain a model that will improve monetization by at least 10%. The end result of the project will be the implementation of innovative technology developed in the course of R&D work in the form of a set of tools for simulating the game economy without the need to use historical player data, to support the game designer in selecting optimal game parameters and to generate personalized offers. The developed technology will combine techniques for storing and analyzing large data sets and machine learning.

Requirements: Python, Machine learning
Additional: Flat structure, Small teams.