Prompt Orchestration Markup Language (POML) (microsoft.github.io)

POML (Prompt Orchestration Markup Language) is a new, HTML-like markup for structuring and managing prompts for large language models. It introduces semantic components such as <role>, <task>, and <example>, plus specialized data tags (<document>, <table>, <img>) to embed or reference external files. A CSS-like <stylesheet> system decouples content from presentation (verbosity, syntax formats), and a built-in templating engine supports {{variables}}, loops (for), conditionals (if), and <let> bindings to generate dynamic, data-driven prompts. Developer tooling includes a VS Code extension (syntax highlighting, auto-complete, hover docs, live preview, inline diagnostics, interactive testing) and SDKs for Node.js (TS/JS) and Python for easy integration into LLM frameworks like LangChain. For the AI/ML community this matters because it standardizes prompt design, improves maintainability and reuse, and reduces format sensitivity that often causes brittle LLM behavior. By formalizing data embedding and presentation controls, POML makes complex, data-rich prompt pipelines reproducible and testable, and enables iterative format+content optimization (cited arXiv papers). Early ecosystem pieces (Rust and Ruby renderers, a Python+Angular chatbot demo) and a growing community suggest cross-platform adoption. Practically, POML can speed productionization of LLM apps, simplify collaboration between engineers and prompt authors, and enable tooling-driven validation and optimization of prompts in ML pipelines.