Very easy to prototype but can get

This is very accurate, for external dev work it seems better to stick to building small scale MVPs

Curious what aspects of it are complicated for production? I'm a relatively new dev so I haven't encountered issues. @mr.niko.la

Would love to hear your thoughts too @Teemu

The only semi complicated thing would be to set up a vector store so that it automatically updates from all data sources as new raw data comes in, adds all the appropriate filtering/meta data, and make sure it doesn't break the RAG eval metrics. I'm probably missing stuff though

Yes it has never been easier to do quick and dirty MVP with minimal startup cost.

Show great results in the prototype stage and using weekly meets with end users to. Build off that mvp next features.

Ideally the best way to build modern apps not just Llm apps

Nothing is overly complicated but rag architecture itself is fairly knew and there is lots of variables to tweak. Furthermore every week can feel like a year advancement so lots of things that are complicated by a few week will be abstracted away. So the core app you build might be not optimal in a few weeks and all that work needs to be refactored.

It’s not a waste of time as you can learn from it how it all works but also its time gone.

At least that is my experience so far.

How is everyone else approaching it ?