This is a very fun idea, and seems very useful. But just from a glance it doesn't seem the most user friendly.

I think it would also be helpful to have something of a guide for the first (few) case(s). I could have missed it if you have one, but the AI hit me like a truck. Turns out it is more difficult to BS an AI that is only designed to do one thing than a regular LLM...

I like the concept, and I think that as long as there is an explanation of what works and why, it could be an interesting learning tool.