Validating output when using with_schema

[internethostage]

Thank you very much for this great gem, it's been a joy to use. And now with version 1.4 I was able to simplify my code considerably by using with_schema.

One thing I'm still doing, though I'm not sure it's necessary anymore, is to validate the LLM's response against the schema I gave it.

My current flow is:

1. Define the schema.
2. Send request with that schema.
3. Validate the returned JSON against the same schema.

Is this last step still needed? I haven't tested the new version enough to be fully confident that the LLM will respect the instructions given (LLM's seem to love to throw a curve ball once in a while).

I would love to hear how others are handling this.

[crmne]

Structured Output ensures that the schema is followed: https://platform.openai.com/docs/guides/structured-outputs?api-mode=chat

If you're paranoid about it, you can always use a JSON Schema validator. Plenty of gems around for that.

[internethostage]

Thanks for the turbo feedback @crmne !

Opened this question mostly to hear from others experience with structured outputs and if they have seen anything funky.
For context I am using gemini, though I like to presume they also validate on the way out, but hard to be fully confident with anything LLM related at the moment.

For defining/validating schema I am using dry-schema and its working great for my purposes.

Thanks once again