Agentic FizzBuzz

A distributed, agent-based FizzBuzz implementation where each microservice operates as an autonomous agent with its own tooling.

Demo Link

Architecture

This system consists of 5 specialized agents, each handling a specific aspect of the FizzBuzz problem:

• Number Agent (Port 8005): Returns the number as a string if not divisible by 3 or 5
• Fizz Agent (Port 8001): Returns "Fizz" if divisible by 3, null otherwise
• Buzz Agent (Port 8002): Returns "Buzz" if divisible by 5, null otherwise
• FizzBuzz Agent (Port 8003): Returns "FizzBuzz" if divisible by both 3 and 5, null otherwise
• Coordinator Agent (Port 8004): Orchestrates the workflow, routes requests to appropriate agents, and returns the final result

How It Works

1. Coordinator Agent receives a number and task via /process endpoint
2. Coordinator validates the task is "fizzbuzz" and number is in range 1-100
3. Coordinator determines which agent to route to based on divisibility:
   • Divisible by both 3 and 5 → FizzBuzz Agent
   • Divisible by 5 → Buzz Agent
   • Divisible by 3 → Fizz Agent
   • Not divisible by 3 or 5 → Number Agent
4. Selected Agent processes the number and returns the result
5. Coordinator returns the final result

Setup

⚠️ Each agent has its own API key from OpenAI - You will need your own account and API keys to be able to run this solution locally.

Docker (Recommended)

docker-compose up --build

Local Development

bun install
bun run dev

API Endpoints

Coordinator Agent (Main Entry Point)

• POST /process - Submit a number for FizzBuzz processing

  {
    "task": "fizzbuzz",
    "number": 15
  }

Individual Agents

• GET /health - Agent health status
• POST /check - Check divisibility (agent-specific logic)

Development

Each agent operates independently and can be scaled horizontally. The system provides:

• Autonomy: Each agent operates independently
• Specialization: Agents focus on specific divisibility checks
• Scalability: Individual agents can be scaled based on load
• Resilience: System continues operating even if some agents fail