Show HN: We cut RAG latency ~2× by switching embedding model | Not Hacker News!