Write pseudocode for a ReAct-style loop
Company: Zillow
Role: Machine Learning Engineer
Category: Coding & Algorithms
Difficulty: medium
Interview Round: Technical Screen
## Coding prompt (pseudocode)
Write **pseudocode** (does not need to compile) for a **ReAct-style agent loop** that alternates between **reasoning** and **actions**.
### Requirements
- The agent receives a **user question**.
- On each iteration, the model produces something like:
- `Thought`: internal reasoning about what to do next
- `Action`: the name of a tool to call (e.g., `Search`, `Calculator`, `LookupDB`) or `Finish`
- `ActionInput`: arguments for the tool
- If an `Action` is chosen:
- Call the tool, obtain an `Observation`.
- Append `Observation` back into the context for the next iteration.
- Stop when the model chooses `Finish` or when a maximum number of steps is reached.
- Include basic safeguards:
- max steps / timeouts
- handling tool errors
- preventing infinite loops
### Deliverable
Provide the loop structure and the data you keep in state (e.g., conversation history, scratchpad, tool outputs).
Quick Answer: This question evaluates the ability to design a ReAct-style agent control loop and related competencies such as state management, tool orchestration, observation handling, error management, and safeguards for termination and resource limits.