LLM Council" (multiple models via OpenRouter), has them review each other's answers, and synthesizes a final response using a Chairman LLM. Based on https://github.com/karpathy/llm-council