CompactRAG: Reducing LLM Calls and Token Overhead in Multi-Hop Question Answering | Signal Canvas | ScienceToStartup