Parse and Validate LLM JSON | Dataford Interview Questions

Write a function to parse and validate a complex JSON output generated by an LLM.

Implement parse_and_validate(raw_output) to return the parsed object when it follows the specified schema, or None for invalid JSON or invalid data. The top-level object must contain exactly answer, confidence, citations, and actions. Validate nested citation and action objects, including action-specific parameter requirements. Reject duplicate keys, unknown fields, incorrect types, out-of-range values, and non-finite numbers.

Function signature: def parse_and_validate(raw_output):

The input is a JSON string. The output is either a Python dictionary containing the validated data or None.

Constraints

raw_output is a JSON string.
The root value must be a JSON object.
answer is a nonempty string of at most 2000 characters.
confidence is a finite number in the range [0, 1].
There are at most 20 citations and 10 actions.
Citation pages are positive integers.
Each action is either retrieve or summarize with the exact parameter schema defined above.

Function Signature

def parse_and_validate(raw_output):

Your question is Parse and Validate LLM JSON. Start with the requirements in the Question tab.

Run and submit as often as you like. When you're ready, talk me through your approach or go straight to the code.

Problem