The AI Interview - Master AI/ML Interviews

Parse function/tool calls from LLM output that may be in different formats. Your parser should:

Detect JSON format and extract function name + arguments
Detect XML-like tags
Parse natural language patterns like 'Call X with Y=Z'
Return confidence score based on parsing certainty

Constraints:

Handle malformed inputs gracefully
Return None for unparseable outputs
Arguments should always be a dictionary
Confidence: 1.0 for JSON, 0.8 for XML, 0.6 for NL

Examples

Example 1:

Input: parse_function_call('{"function": "search", "arguments": {"query": "python"}}')

Output: {'function': 'search', 'arguments': {'query': 'python'}, 'confidence': 1.0}

Explanation: Valid JSON format detected, full confidence

Example 2:

Input: parse_function_call('Call calculator with x=5, y=10')

Output: {'function': 'calculator', 'arguments': {'x': '5', 'y': '10'}, 'confidence': 0.6}

Explanation: Natural language pattern matched, lower confidence due to string values

Starter Code

import json
import re

def parse_function_call(llm_output):
    """
    Parse function calls from LLM output.
    Handles both JSON and natural language formats.
    
    Expected formats:
    1. JSON: {"function": "name", "arguments": {...}}
    2. XML: <function>name</function><args>{...}</args>
    3. NL: Call function_name with arg1=value1
    
    Returns: dict with 'function', 'arguments', 'confidence' (0-1)
    """
    # Your implementation here
    pass

Function Call Parser

Examples

Starter Code