ai-structured-extraction.yaml

name: ai-structured-extraction
description: >
  Fetch a webpage and use an LLM with output_schema to extract structured data
  (title, author, key topics). Demonstrates enforcing JSON structure from AI output.

inputs:
  url:
    type: string
    description: URL of the page to fetch and extract data from

steps:
  - name: fetch-page
    action: http/request
    timeout: "15s"
    retry:
      max_attempts: 2
      backoff: exponential
    params:
      method: GET
      url: "{{ inputs.url }}"

  - name: extract-metadata
    action: ai/completion
    credential: openai
    params:
      model: gpt-4o
      system_prompt: >
        You are a structured data extraction engine. Given raw page content,
        extract the requested fields accurately. If a field cannot be determined,
        use null or an empty value as appropriate.
      prompt: >
        Extract the following metadata from this page content:

        {{ steps['fetch-page'].output.body }}
      output_schema:
        type: object
        properties:
          title:
            type: string
          author:
            type: string
          key_topics:
            type: array
            items:
              type: string
        required:
          - title
          - author
          - key_topics
        additionalProperties: false

  - name: post-extracted-data
    action: http/request
    if: "steps['extract-metadata'].output.json.title != ''"
    params:
      method: POST
      url: "https://jsonplaceholder.typicode.com/posts"
      headers:
        Content-Type: "application/json"
      body:
        title: "{{ steps['extract-metadata'].output.json.title }}"
        author: "{{ steps['extract-metadata'].output.json.author }}"
        body: "Extracted structured metadata from {{ inputs.url }}"