For ROAI Prelot, I had the chance to contribute from a different perspective: not only as someone who solves problems, but also as a volunteer and problem setter. This helped me understand how much work goes into creating a good contest task.
My proposed problem was called Funtime Beach. It was a Reinforcement Learning task inspired by Gymnasium's Blackjack environment, but reinterpreted in a more original and playful setting. The goal was to keep the decision-making feeling of Blackjack while turning it into a model-based RL problem.
Designing the task was challenging because a good problem needs to be clear, fair, and interesting at the same time. I had to think about how participants would understand the environment, what strategies they could discover, and how the problem statement should guide them without giving away the solution.
This experience helped me improve a lot. It made me better at explaining technical ideas, thinking about edge cases, and understanding competitions from the organizer side. It also made me appreciate problem setters much more, because creating a task can be just as difficult as solving one.
