Convergence Take-Home Test

🚀 Your Task

Your goal is to build a Deep Research agent that can answer questions about Pokemon. The agent should be able to reason through the task, use tools to gather information, and return a useful response.

🧭 Expectations

Your submission should include:

1. A simple command line client or working application.
2. An agent you implemented yourself (not just calling an agent library).
3. Well-organised code that is easy to review.
4. A README with instructions on how to set up and run your agent.

Agent Design

The agent should:

1. Reason and plan how it will solve the user's query.
2. Gather information from the Pokémon API, calling tools multiple times as needed.
3. Conduct multiple rounds of thinking and acting as needed.

We recommend you start simple and slowly expand your agent.

Your README should outline your final agent design, including how the agent solves problems and what sort of tools it has.

Evaluation

We recommend you first get things flowing end-to-end (your agent can reason and call tools). Then it's time to iterate on your agent.

Start out by trying a few simple queries.

If things don't work yet, try even simpler (break the tasks down even more or add extra information in your query: "Use the get_pokemon tool to return data about Pikachu").

Slowly improve its outputs, perhaps by adding new tools, fixing existing tools, or improving your system prompts.

Make sure to play around lots with your agent and actually read its outputs. If you don't know much about Pokemon, you can ask ChatGPT to evaluate the response (although it sometimes mixes up knowledge across games).

---

Using the Template

Feel free to start fresh and use your own setup.

Prerequisites

• Docker

Development Setup

1. Build and run the development environment:

   make build

   then

   make start

2. Add dependencies once in docker container: Add new libraries in requirements/dev.in

   then

   make deps

Default Project Structure

Feel free to adjust.

convergence-interview/
├── src/                    # Source code with agent code
│   └── main.py             # Main application entry point (CLI)
├── tests/                  # Test files
├── requirements/           # Python dependencies
├── Dockerfile              # Docker image configuration
├── docker-compose.dev.yaml # Development Docker Compose
└── Makefile                # Common development commands

There is no expectation to write tests. It might help to isolate API behaviour to ensure the tools work correctly.