Could anyone provide more details on how files uploaded to an assistant's knowledge base are processed and utilised? Specifically, does Vapi implement a RAG system with vector embeddings and semantic search, or is there a different method for integrating this information into the assistant's responses?