A few weeks ago, during the LK-99 hype, I made a demo app to explain superconductors in simple terms using RAG over relevant Wikipedia pages. I thought it would be cool to extend this to all physics-related Wikipedia pages (which turns out to be ~ 14K). After getting the text and splitting the pages, I ended up with around 100K chunks. I created all the embeddings using OpenAI’s embedding API, which cost around $7. I stored all the vectors in Pinecone.