Training Box: Web-Based Fighting Game Execution Trainer

• 

  the config screen (at least the controller detection works lol)

Inspiration

This project is an unfinished OpenAI hackathon entry. I ran out of time before the submission deadline, but I learned a great deal and still intend to finish it.

What it does

It detects your controller in the browser and then walks you through exercises to improve your fighting game execution execution.

How we built it

5.6 Sol Low and Terra Low inside oh my pi

Challenges we ran into

The biggest challenge was that I used the Grill Me with Docs skill and let the resulting lessons steer the project too far. In practice, those lessons were not useful enough to justify the time they consumed. In retrospect, I should have rolled back to the earlier commit where the web server worked and reliably detected the gamepad, used GPT Image 2 to generate a leverless-controller layout and button icons, and built roughly six basic Input Primitive exercises.

Instead, I became indecisive about whether the trainer should run on the RP2040 itself. I spent too long weighing a TypeScript implementation constrained to roughly 500 KB against using a game engine to build a simple sprite-based experience. I eventually found a promising engine, Phaser, and a suitable project to fork, but I was called into work on the final day and could not complete that direction.

Although I missed the deadline, I am grateful to OpenAI and Devpost for the opportunity to participate. The project will continue after the hackathon.

Accomplishments that we're proud of

What we learned

What's next for Training Box: Web-Based Fighting Game Execution Trainer

Built With

• gpt5.6-sol