It kind of is a bit of both. Finding a large model isn’t too hard. I mean Gemini 1.5 Pro eats movie scripts and finds the right answer in there or in huge codebases.

I implemented a RAG LLM app that runs locally on my machine with a local model and it scans my source code directories and gives me answers about them. And it was less than 100 lines of Python.

But productionising it and making it run at scale is a whole other topic which I’m not too familiar with yet