llm api structured json

If you want an LLM API to return structured JSON, the common approaches are:

1) Use a JSON-only instruction

Ask the model to output only valid JSON.

Prompt example

Return only valid JSON with these fields:
- name: string
- age: integer
- skills: array of strings

Expected output

{
  "name": "Alice",
  "age": 29,
  "skills": ["Python", "SQL"]
}

2) Use schema-based output

If your API supports it, provide a schema so the model must conform to it.

Example JSON Schema

{
  "type": "object",
  "properties": {
    "name": { "type": "string" },
    "age": { "type": "integer" },
    "skills": {
      "type": "array",
      "items": { "type": "string" }
    }
  },
  "required": ["name", "age", "skills"],
  "additionalProperties": false
}

3) Validate after generation

Even with structured output, you should still:

• parse JSON
• validate against schema
• retry if invalid

4) Use function/tool calling

Many LLM APIs support function calling or tools, which is often better than raw JSON prompting.

Example

{
  "name": "create_profile",
  "arguments": {
    "name": "Alice",
    "age": 29,
    "skills": ["Python", "SQL"]
  }
}

If you want, I can give you:

• a Python example
• an OpenAI API example
• a JSON schema template
• or a best-practice prompt for structured output.