So I’ve gotten a lot of progress done recently on the Vantage fit project. So much so that I’m almost done with the meat and potatoes of the exercise builder and runner and all of those compoents of a basic, compoetent, training app. Therefore, I’ve decided to begin working on the RAG pipeline.
I’ve just been recently trying to read up on how RAG pipelines tend to be built today. I built a very basic one last year but a lot of the technology has obviously changed so I’ve been refamilizaring myself.
I followed a particular tutorial from huggingface and it’s definitley much more straight forward to build out a RAG pipeline compared to last year I feel. It was super easy to load a basic model from llama and vector database, store the corpus and then prompt for a really good answer based on the corpus.
Thankfully, last year in particular I wrote out what will be the corpus for our RAG pipeline that will be used in Vantage fit. I need to obviosuly write out some more but I’ve got dozens of pages already of instructions on how to resistance train and all of these things.
So yeah, I’m going to play around with the RAG some more and see how it reacts with my corpus, I’ll look into other methods of determining similarity other than just the cosine similarity and see how that works out.