The AI Interview - Master AI/ML Interviews

Deterministic vs Non-Deterministic Tool Usage

Some tools always produce the same output for the same inputs (deterministic); others do not.

Task

Build ToolClassifier that:

Registers tools with a determinism flag.
Caches results of deterministic tools by content-hash.
Always re-executes non-deterministic tools.
Returns {'result': Any, 'cached': bool, 'deterministic': bool}.

Constraints

Non-deterministic tools NEVER use cache.
Cache key: sha256(JSON of name + sorted kwargs).
Calling an unknown tool raises KeyError.

Examples

Example 1:

Input:

tc.register('add', lambda **k: k['a']+k['b'], True)
tc.call('add', a=2, b=3)
tc.call('add', a=2, b=3)

Output: {'result': 5, 'cached': True, 'deterministic': True}

Explanation: Second call with same args returns from cache.

Starter Code

from typing import Any, Dict, Callable
import hashlib, json

class ToolClassifier:
    def __init__(self):
        self.registry: Dict[str, dict] = {}
        self.cache: Dict[str, Any] = {}

    def register(self, name: str, fn: Callable, deterministic: bool) -> None:
        pass

    def call(self, name: str, **kwargs) -> Dict:
        # Returns {'result': Any, 'cached': bool, 'deterministic': bool}
        pass

    def _cache_key(self, name: str, kwargs: Dict) -> str:
        data = json.dumps({'tool': name, 'args': kwargs}, sort_keys=True)
        return hashlib.sha256(data.encode()).hexdigest()