Multi-agent debate simulator using RAG to let LLM-powered agents debate with personality and knowledge-grounding