Valerie's Tech Writer Garden

POML

2 min read

Prompt Orchestration Markup Language (POML) is a markup language that offers structure, readability, modularity, and maintainability for AI prompt engineering.

Tip

POML is best used in combination with Retrieval Augmented Generation (RAG).

Resources

Key Features

  • Semantic components:
    • <role>
    • <task>
    • <example> 
  • Data components:
    • <document>
    • <table>
    • <img>
  • Styling (e.g., verbosity, syntax format):
    • <stylesheet>
  • Javascript expressions:
    • Variables: {{variableName}}
    • Arithmetic: {{a + b}}, {{x * y}}, {{count / total}}
    • String Concatenation: {{firstName + " " + lastName}}
    • Array Access: {{myArray[0]}}
    • Object Property Access: {{myObject.propertyName}}
    • Function Calls: {{myFunction(arg1, arg2)}} (if myFunction is defined in the context)
    • Ternary Operators: {{condition ? valueIfTrue : valueIfFalse}}
    • Accessing loop variables: {{loop.index}} (explained in the “For Attribute” section)

Examples

General Use

<poml>
  <role>You are a patient teacher explaining concepts to a 10-year-old.</role>
  <task>Explain the concept of photosynthesis using the provided image as a reference.</task>
 
  <img src="photosynthesis_diagram.png" alt="Diagram of photosynthesis" />
 
  <output-format>
    Keep the explanation simple, engaging, and under 100 words.
    Start with "Hey there, future scientist!".
  </output-format>
</poml>

Import Data from a File

<poml>
  <let name="users" src="users.json" />
  <p>First user: {{users[0].name}}</p>
</poml>

Style Class

<poml>
  <table className="csv" records="[[1,2,3],[4,5,6]]"/>
  <stylesheet>
    {
      ".csv": {
        "syntax": "csv",
        "writerOptions": "{\\"csvSeparator\\": \\";\\", \\"csvHeader\\": false}"
      }
    }
  </stylesheet>
</poml>

Combination with RAG

In a typical RAG pipeline, you have:

  1. User Query → Retriever
    • Vector search or hybrid search pulls top-K documents from your knowledge base.
  2. *Context Assembly
    • Normally, devs concatenate retrieved chunks into a giant string prompt.
    • With POML, you can structure this step cleanly.
  3. LLM Prompt Execution
    • POML renders a prompt with <role>, <task>, <document>, etc., which feeds into the LLM.

Without POML:

You are a helpful assistant. Answer the question.
Documents:
[doc1] ...
[doc2] ...
Question: What are the side effects of Drug X?

With POML:

<poml>
  <role>You are a medical assistant providing safe, factual answers.</role>
  <task>
    Based on the provided documents, summarize the side effects of Drug X.
    If unsure, say "I don’t know."
  </task>
 
  <context>
    <documents>
      <document id="doc1" source="PubMed"> 
        {{ retrieved_doc_1 }}
      </document>
      <document id="doc2" source="FDA">
        {{ retrieved_doc_2 }}
      </document>
    </documents>
  </context>
 
  <output-format>
    List common side effects in bullet points. 
    Keep under 150 words.
  </output-format>
</poml>

Graph View

Backlinks