Thank you for checking it out!
I'll check out the first guy, he was supposed to be a very easy solve, with an answer appearing as a message 5 inputs in, but he shouldn't be confessing right away.
The LLM is a realistic idea, but generating decent audio dynamically isn't. I'm still testing solution that's supposed to be the middle ground, but the current one could work with larger amount of data.
Thanks for giving it a try, I do appreciate it.