🧾 Designing the Output Schema

The assistant produces structured outputs for each job posting based on a customizable schema defined in default_schema.json. This file tells the assistant what fields to return and how to format them.

📂 Where It Lives

Path: flow_jobposting/default_schema.json

This schema is read every time a prompt is executed. Each field defines:

• name: name of the column
• description: what the assistant should extract or infer
• type: data type (string, boolean, number, etc.)

---

✏️ How to Customize

To customize what the assistant outputs:

1. Open default_schema.json
2. Add, remove, or edit a field
3. Save and rerun the prompt pipeline

Example 1 — Add a field

{
  "name": "salary_guess",
  "description": "Estimated annual salary in USD, if mentioned or inferable.",
  "type": "number"
}

Example 2 — Change a field

From:

{
  "name": "recommend_apply",
  "description": "Should the user apply to this job? Answer Yes or No.",
  "type": "string"
}

To:

{
  "name": "recommend_apply",
  "description": "Score the job's quality from 1 (poor) to 5 (excellent).",
  "type": "number"
}

⚠️ Tip: The schema must match the prompt. If you ask for a 1–5 score but keep the Yes/No prompt, results may fail or be inconsistent.

---

✅ Try It Yourself

Start small:

• Add one optional field
• Test it with the pipeline (09_run_promptflow.py)
• View the results in the results tab of the app

If successful, you’ll see your new column appear in the match table and CSV outputs.

---

📈 Why Use This?

• Tailor results to your personal criteria
• Extract deeper insights (culture, visa, location match)
• Power downstream automation (e.g., salary filters)

Schema customization gives you control over what the assistant returns — and how useful it will be for your workflow.