Ran some experiments to see how different things affected the quality of results from a retrieval-augmented generation (RAG) pipeline. Post shows results and details / code for how to run similar experiments yourself.