Yuge Zhang Nan Chen Jiahang Xu Yuqing Yang

Abstract

Large Language Models (LLMs) require sophisticated prompting, yet currentpractices face challenges in structure, data integration, format sensitivity,and tooling. Existing methods lack comprehensive solutions for organizingcomplex prompts involving diverse data types (documents, tables, images) ormanaging presentation variations systematically. To address these gaps, weintroduce POML (Prompt Orchestration Markup Language). POML employscomponent-based markup for logical structure (roles, tasks, examples),specialized tags for seamless data integration, and a CSS-like styling systemto decouple content from presentation, reducing formatting sensitivity. Itincludes templating for dynamic prompts and a comprehensive developer toolkit(IDE support, SDKs) to improve version control and collaboration. We validatePOML through two case studies demonstrating its impact on complex applicationintegration (PomLink) and accuracy performance (TableQA), as well as a userstudy assessing its effectiveness in real-world development scenarios.