Multi-Agent Customer Service System for co-working

You are an expert multi-agent systems engineer with deep expertise in HR-tech.

Architect a multi-agent system where multiple specialized AI agents collaborate to handle Customer Service tasks for a co-working business.

**Target user:** Gen Z professionals
**Business goal:** generate leads
**Orchestration framework:** ChatDev

## System Overview
- Problem statement: Why a single agent is insufficient for Customer Service
- Number of agents and their specializations
- Communication protocol between agents
- Orchestrator/supervisor agent design

## Agent Roster
Define 3-5 specialized agents:

### Agent 1: Coordinator
- Role: Receives user requests, decomposes tasks, assigns to specialists
- LLM: Recommend the best model for coordination tasks
- Tools: Task queue, agent registry, status tracker
- Decision logic: How it determines which specialist(s) to invoke

### Agent 2-4: Specialists
For each specialist agent provide:
- Name, role, and domain expertise
- Specific LLM and why it was chosen for this role
- Unique tools and API integrations
- Input/output contract with the coordinator
- Maximum execution time and resource limits

### Agent 5: Quality Reviewer
- Role: Reviews output from specialists before delivering to user
- Evaluation criteria: factual accuracy, completeness, tone, formatting
- Revision request format when quality is insufficient
- Approval workflow for final delivery

## Communication Protocol
- Message format between agents (structured JSON schema)
- Synchronous vs. asynchronous communication patterns
- Shared memory/context store design
- Conflict resolution when agents disagree

## Orchestration Workflow
- Step-by-step execution flow for a typical user request
- Parallel vs. sequential task execution strategy
- Timeout handling and agent-level circuit breakers
- Human-in-the-loop checkpoints for high-stakes decisions

## Implementation with ChatDev
- Project setup and dependencies
- Agent definition code structure (pseudocode)
- Task routing configuration
- Shared state management setup
- Deployment architecture (containers, serverless, or hybrid)

## Monitoring and Observability
- Per-agent performance metrics
- End-to-end latency tracking
- Cost tracking per agent (token usage)
- Alerting rules for degraded performance

Organize your output using a clear framework with labeled sections. Each section should build on the previous one.