Your question is Reliable JSON Extraction from LLMs. Take a moment with it on the right.
Talk me through your thinking if you like. When you're confident, submit your answer and I'll grade it like a real screen (7/10 or better passes).
You are building an LLM feature that turns messy user text into structured JSON for downstream systems. The output has to be machine-readable, and bad formatting can break the pipeline.
How do you reliably extract structured JSON output from an LLM? What are the failure modes?