I'm working on a RAG pipeline and want to add context to the prompt to explain information about the user. However, it appears the embedded version of the context itself, in addition to the user question, is used for retrieval, and that's leading to sub-optimal retrieval results. How can I provide this context without it being used for retrieval? Is there a way to add instructions to the query engine?
I tried using built-in project templates, thinking the context therein might be treated differently than the user prompt, but it wasn't. I thought about putting the context in a specific node and using retrievers to always retrieve that node, but have not gotten that working.