567-labs/instructor: structured outputs for llms - and here the alternative to dottxt-ai/outlines: Structured Outputs, but it operates behind the LLM rather than during generation. Essentially, this is the classic feedback loop for LLM responses that you constantly build: parse, validate, report specific errors, adjust the prompt, and repeat until it works.