Extract JSON from an LLM Reply

You asked the model for JSON. It replied with "Sure! Here's the data:", then the JSON, then a friendly sign-off. Robust apps extract the JSON instead of crashing.

Implement extract_json(text) that returns the first complete JSON object ({...}) in text, parsed with json.loads, or None if there isn't one.

Rules

• Scan for { and find its matching } by tracking nesting depth.
• Braces inside JSON strings don't count, and strings can contain escaped quotes (\").
• If a candidate isn't valid JSON, keep scanning from the next {.
• The JSON may be wrapped in a Markdown code fence (```json ... ```). Your scan handles that naturally.