You don't need Langchain, here's how to do RAG without it

You mentioned quantizing the vector index to reduce size. I am curious about how much compression were you able to achieve and how did that affect performance?