The AI Interview - Master AI/ML Interviews

ReAct Pattern Implementation

ReAct (Reasoning + Acting) interleaves chain-of-thought reasoning with tool actions.

Task

Implement a ReActAgent that:

Maintains a trace of Thought → Action → Observation steps.
Terminates when a final answer is produced or max_steps is reached.
Supports pluggable tools via a dictionary.

Constraints

Each step must log: thought, action, observation.
If no tool matches, return an error observation.
Return '[Max steps exceeded]' if limit reached.

Examples

Example 1:

Input:

tools = {'add': lambda a, b: a + b}
agent = ReActAgent(tools)
agent.run('What is 2 + 3?')

Output: '5'

Explanation: Agent thinks, calls add(2,3), observes 5, returns answer.

Starter Code

from typing import Dict, Any, Optional

class ReActAgent:
    def __init__(self, tools: Dict[str, callable]):
        self.tools = tools
        self.trace = []

    def think(self, observation: str) -> str:
        # TODO: Produce a thought given observation
        pass

    def act(self, thought: str) -> Dict[str, Any]:
        # TODO: Return {'tool': name, 'args': {...}} or {'answer': '...'}
        pass

    def observe(self, tool: str, args: Dict) -> str:
        # TODO: Execute tool and return observation
        pass

    def run(self, task: str, max_steps: int = 5) -> str:
        # TODO: Implement Reason-Act-Observe loop
        pass

Implement the ReAct Pattern

ReAct Pattern Implementation

Task

Constraints

Examples

Starter Code