Unconstrained method

In the unconstrained method, you send a good prompt, let the LLM produce output without constrained decoding or any other interference, and add code at the end to parse, validate, and repair the LLM output.

Some libraries implement this for us, i.e., you input your schema, and the library writes the prompt and the parse-and-repair code, and uses it to request the LLM output.