Portfolio

Dino - An environment, where agents are able to choose when to act

This is a project where I experiment with efficient decision-making for agents in continuous time environments. I compared the default approach with my custom delayed action system and wrote an article about the results on Medium.

Tools for ML-Agents

Streamlining the process of training and use of techniques similar to Population Based Training. It also allows you to spawn multiple environments, change camera between different agents, adjust the timescale, display the agent related information, and debug the custom values on the Tensorboard while training.

Doodle Jump - Unity platformer with ML Agents

Made to design and compare different perception systems for reinforcement learning agents and try out different training algorithms. I tested the potential of a custom sensor that parses surrounding objects into a variable length vector that is processed by the attention module. It also served as a benchmark to compare the impact of different setups on the decision sampling speed.

Fealty To The King - Unity board game with ML Agents

A unique turn-based board game with a chess-RPG blend of rules. By using reinforcement learning and self-play, I developed and trained an agent to play against a player. The game was the capstone project for the Game Design and Development program. It was made by a team of five, in less than six months.

Unity DOTS ML Agents Prototype

A multi-threaded multi-agent environment where hundreds of agents compete to survive in a battle-royale style game. It is developed using the experimental Unity ML-Agents Data-Oriented Technology Stack toolkit. The research was part of my volunteer work at Enhearten Media to run the stability tests and see what could be achieved by using per-agent AI in an RTS game. In an adversarial environment, two networks were trained to see if any interesting behavior emerges.

Endless runner “Star Jump Commando”

An endless runner game Star Jump Commando for Android & IOS. The game was released in April 2022.

Unity Developer / Programmer

• Developed and implemented collectibles and equipment systems, enemies, bosses, wave generation system and difficulty multiplier.
• Collaborated with other programmers to create a general code architecture and support existing code systems.
• QA, outlined the balancing direction and adjusted the difficulty of the game’s progressing.
• Tightly worked with a design team to implement their game vision and outlined tasks during agile processes and standups.