Repair Malformed JSON from an AI Response

LLMs frequently return invalid JSON despite being instructed to produce valid JSON output. Common failure modes include trailing commas after the last object property, single-quoted strings instead of double-quoted, unquoted keys, Python-style None and True/False instead of null and true/false, and truncated JSON when the response hits the max_tokens limit. This example shows a realistic malformed AI response and demonstrates how the repair tool fixes each issue automatically. The repair tool applies a sequence of targeted transformations: it converts single quotes to double quotes while preserving apostrophes in string values, removes trailing commas before closing brackets and braces, replaces None/True/False with their JSON equivalents, and attempts to close unclosed brackets and braces when truncation is detected. Each transformation is applied in order because fixing one issue (like single quotes) can expose the next (like an unquoted key). For production AI pipelines, JSON repair should be the first step in your response parsing logic before JSON.parse(). If repair still fails to produce valid JSON, fall back to extracting the partial data with a lenient parser or prompting the model again with the specific error. Never crash the pipeline because the model returned imperfect JSON.

{
  'name': 'Alice Johnson',
  'age': 32,
  'active': True,
  'score': None,
  'tags': ['ai', 'developer', 'python',],
  'address': {
    'city': 'San Francisco',
    'zip': '94105',
  },
}