A massive Tetris fan, Ben van Oostendorp originally arrived at DigiPen with dreams of creating the next hit game as a BA in Game Design major. While his love for Tetris persisted through school, eventually leading him to start the DigiPen Tetris Club, his professional interests rotated in a new direction after his freshman year. That’s when he began to develop a keen interest in artificial intelligence. “I started really enjoying programming and math, and I was fascinated by genetic algorithms,” van Oostendorp says, referring to a computer science optimization technique inspired by biological evolution. “After learning about them, I stumbled into basic classification problems and eventually simple neural networks. At that point I was pretty hooked.”

Honoring his budding new passion for AI models, he transferred to the BS in Computer Science in Artificial Intelligence program (formerly the BS in Computer Science in Machine Learning). That path eventually led to his current role as an Amazon software developer, but not before a few fun stops along the way applying his growing AI skillset to games.
For his junior year, van Oostendorp got to showcase just how far he had dug into AI with his self-playing Shovel Knight project. Using the popular 2D platformer game as a training environment, van Oostendorp developed a custom engine and AI agent that learned how to make its way through levels on its own. “I used a relatively novel technique of object detection and bounding boxes to feed into a powerful reinforcement learning agent, then applied this same mechanism to other environments to prove its effectiveness,” van Oostendorp says. The longer the AI agent played the game, the better it got at defeating enemies, scaling platforms, busting through blocks, and making its way to finish.
During his senior year, van Oostendorp took one of his AI research projects across the country to Cincinnati, Ohio, for the 2023 North American Fuzzy Information Processing Society (NAFIPS) annual conference. “This was one of my favorite things I did at DigiPen!” van Oostendorp says. The conference focused on research in “fuzzy logic,” a rules-based reasoning model that goes beyond binary “yes” and “no” decision making by allowing for shades of grey in between. The model is especially useful in developing “explainable” AI, where users don’t have to guess how the AI’s decisions were made — they can actually read how it reached its conclusions, an especially crucial dimension in high-stakes AI applications like medicine and the military.

For his part of the project, van Oostendorp helped train an AI agent to play a game called Cart Pole — a classic challenge used to test AI that tasks it with balancing a pole on a moving cart. He worked on a method that combined traditional neural networks with fuzzy logic to show how the AI could not only learn the task, but also explain how and why it made certain decisions. “Our explainability mechanism was using an adaptive neuro-fuzzy inference system applied to neural networks for training, which provided novel and useful explanations of what the neural network was actually doing and ‘seeing,’” van Oostendorp says of the technique, which he presented on a panel at NAFIPS.
Shortly after graduating, van Oostendorp’s AI portfolio impressed Amazon, who hired him on as a software developer in 2023. Lately, he’s been putting his talents to work on front-end development, focusing on the user-facing aspects of Amazon’s offerings. “I got my first taste of working on the front end of our application and was hooked immediately,” van Oostendorp says. “I really enjoyed the process of making new and user-friendly experiences, as well as updating previously existing features and leveling them up.”
While the skillsets are sometimes different, van Oostendorp says the framework for developing and optimizing AI set him up for success in his front-end work today. “Machine learning is a very iterative field, constantly making small updates or tweaks, and that transferred extremely well into front-end development!” he says.
The deep note-taking skills he developed at DigiPen have also been critical to his work at Amazon, allowing him to go back and examine his own logical reasoning much like the explainable AI models he wrote as a student. “Being able to write notes for myself so that I can come back to any problem or question I was working on and catch back up or re-polish the knowledge I had has allowed me to iterate quickly and provide valuable context, when needed, to myself and others,” van Oostendorp says.