RAG 检索增强生成原理图

生成一张 4:5 竖版「RAG 检索增强生成原理图」案例图，所属分类为「信息图与示意图」。瑞士风格平面信息图，展示五步 RAG 流程，海军蓝与珊瑚色节点，细线箭头，浅黄色引用标注框。画面需要完整呈现上述主体、构图、配色、材质、光线和整体风格；可见文字以自然简体中文为主，必要的技术缩写、代码、变量名、学术符号、单位、拉丁物种名或装饰性外文可保留。

A clean Swiss-style technical diagram explaining Retrieval-Augmented Generation (RAG) on a warm off-white A4 canvas. Five numbered steps flow left to right: a user query icon, a vector database cylinder labeled "Vector Store", a retrieval arrow labeled "Top-K chunks", a language model rectangle labeled "LLM", and a final answer speech bubble. Each node is a rounded rectangle in navy, slate, or coral. Thin hairline connectors with directional arrowheads. Callout box in pale yellow reads: "RAG grounds the model in real facts, not just training memory." Title set in a bold geometric sans at 28pt. Sub-labels in 11pt light weight. Grid-aligned layout, 12-column structure. No gradients — flat fills only. Footer in small caps: "Architecture Overview". White background, generous margins.