📄 中文摘要
在当今的 AI 应用中，常常局限于 RAG（检索增强生成）或提示工程的二元对立。然而，在构建我的多人 lateral thinking puzzle 游戏 TurtleNoir 时，我意识到仅靠实时推理是不够的。为了创造一个真正具有“灵魂”的 AI 主持人，我采用了双层架构设计。这种设计不仅提升了 AI 的互动性，还增强了其在复杂推理情境中的表现。通过这种方法，AI 主持人能够更有效地引导玩家进行 lateral thinking，提供更丰富的游戏体验。
超越 RAG：构建具有逻辑基础的双层架构 AI 主持人
📄 English Summary
Beyond RAG: Building a Logic-Grounded AI Host with a Dual-Layer Architecture
Current AI applications often get stuck in the binary of Retrieval-Augmented Generation (RAG) or Prompt Engineering. However, while developing TurtleNoir, a multiplayer lateral thinking puzzle game, it became clear that real-time inference alone was insufficient. To create an AI Host with a true 'soul,' a dual-layer architecture was implemented. This design not only enhanced the interactivity of the AI but also improved its performance in complex reasoning scenarios. As a result, the AI Host can effectively guide players through lateral thinking, providing a richer gaming experience.