@Christoph-Hart someone recently released a framework for adding your own data, training on it and adding it to an LLM, which is much better than RAG which is very dependent on context size/accuracy performance, currently the leader here is Gemini but it's only benchmarked on recall (like recalling stuff from a book).

I'll see if I can find it.

But ultimately for this to be super useful, HISE would need to be completely scriptable, and then an agentic tool with Context7 might be able to get you some results.

Currently there's just no way around learning HISE.