LLM returns JSON wrapped in ```json fences and json.loads fails
@devsingh05▲ 13 net votes
Models often wrap structured output in a markdown code fence or add a sentence before it, so parsing raises 'Expecting value: line 1 column 1'. Ask for JSON only and still parse defensively: strip a leading and trailing fence, or take the substring from the first { to the last }. Better, use the provider's structured-output or tool-calling feature so the SDK returns a parsed object, then validate it with a schema and retry once on failure.