I built a no-code tool that learns to play any 2D game by watching you play it

I've been building a desktop app called DeepEpoch for a while and finally put it on the Microsoft Store. The idea: instead of scripting a game bot or writing training code, you just play the game yourself. The app records your screen and your key inputs together, and trains a model via behaviour cloning to play the way you did. Then you can watch it play — and when it messes up, you take over for a second, it records your correction, and keeps learning. Human-in- the-loop fine-tuning, no code at all. Most stuff in this space is either research repos you have to wire up yourself, or dumb macro recorders that just replay the same clicks. I wanted the middle: real imitation learning, but usable by someone who doesn't want to touch Python. You can train locally if you have a GPU, or in the cloud. It's early and rough in places, but the core loop works. Genuinely curious what people here think — especially whether the "play the game to teach the AI" approach makes sense to you, or where you'd expect it to break. Link in the comments.