Yes, you're right. JavaScript would be the natural solution and I've seen others and had my scikit-learn models in a production environment that way. Although, I'm not too experienced with deploying ML solutions through JS as this was done by a software engineer at the place I worked at before.

I'll definitely consider building something in JS if it comes to building part 2 of this article.

As for the idea of HAL, that is very interesting! I don't imagine this bot going like that as it is built with a small model and it's also trained on Socrates. It can get "annoying" with the Socratic method and returning the question back to you though. :D