Multi-Agent System

Blue → Purple Belt Multi-day project

ภาพรวม

สร้าง multi-agent system ที่แก้ปัญหาซับซ้อนผ่าน agent collaboration This capstone tests your understanding of agentic workflows, tool use, and multi-agent coordination.

Prerequisites

Current Belt: Blue
Quests Completed: At least 50 quests
Modules Covered: Module 1-5

ข้อกำหนด

Phase 1: Design (Day 1-2)

Design agent architecture (manager-worker, peer-to-peer, or hierarchical)
Define agent roles and responsibilities
Design tool interfaces for agent capabilities
Create interaction diagrams showing agent communication

Phase 2: Implementation (Day 3-5)

Implement at least 3 specialized agents
Build a coordinator/orchestrator agent
Implement tool registry and dynamic tool loading
Add inter-agent communication (message passing or shared state)
Implement error handling and fallback strategies

Phase 3: Testing (Day 6)

Unit tests for individual agents
Integration tests for agent interactions
Load testing with multiple concurrent tasks
Failure scenario testing (agent crashes, tool failures)

Phase 4: Polish & Deploy (Day 7)

Add logging and observability
Create monitoring dashboard
Deploy with proper documentation
Write architecture documentation

Example Projects

Choose one or propose your own:

Code Review Bot - Multiple agents reviewing code from different angles (security, performance, style)
Research Assistant - Agents that search, analyze, and synthesize information
Task Automation - Agents that break down complex tasks and execute them
Data Pipeline - Agents that extract, transform, and load data from multiple sources

Rubric

Criteria	Weight	Description
Architecture	30%	Clean agent design, proper separation, extensible
Agent Quality	25%	Agents are focused, effective, and handle errors
Tool Integration	20%	Tools are well-designed and properly used
Testing	15%	Comprehensive test coverage for agents and tools
Documentation	10%	Clear docs on architecture and usage

Getting Started

npx bluebeltdojo download capstone-2-multi-agent
cd capstone-2-multi-agent

Key Concepts to Demonstrate

Agent Loop - Observe → Think → Act cycles
Tool Use - Dynamic tool selection and execution
Multi-Agent Coordination - Agents working together
Error Handling - Graceful degradation when agents fail
Observability - Logging and monitoring agent behavior

Tips

Start with a simple 2-agent system and expand
Use the patterns from Quest 65 (Multi-Agent Debate) and Quest 66 (Hierarchical Agent Team)
Implement proper logging from the start
Test failure scenarios - agents WILL fail
Document your design decisions in ADRs