Skip to Content
WorkshopsWorkshop: Multi-Agent Service

Workshop 03: Multi-Agent AI Service — Supervisor + Specialist Agents

Duration: 120 minutes | Level: Advanced | Audience: AI Engineers, Solution Architects

Learning Objectives

  1. Design a supervisor + specialist agent architecture
  2. Configure agent routing, handoff protocols, and shared state
  3. Implement guardrails for multi-agent safety (max hops, audit trail)
  4. Run evaluation against quality thresholds

Prerequisites

  • Azure subscription with Azure OpenAI access
  • Node.js 18+ or Python 3.10+
  • VS Code with FrootAI extension

Workshop Flow

Part 1: Scaffold Multi-Agent Play (10 min)

npx frootai scaffold 07-multi-agent-service
cd 07-multi-agent-service
code .

Part 2: Review Supervisor Config (20 min)

  • Open config/openai.json — examine supervisor routing rules and specialist agent configs
  • Open config/guardrails.json — review max 5 hops, 30s timeout, audit trail settings

Part 3: Build the Supervisor (40 min)

Open Copilot Chat:

@builder Build a supervisor agent that routes requests to research_agent, action_agent, and analysis_agent based on the user intent

Part 4: Build Specialist Agents (30 min)

@builder Build the research_agent that answers questions using the FrootAI knowledge base

Part 5: Test & Evaluate (20 min)

@tuner Run evaluation with scores: groundedness=4.2, relevance=3.8, coherence=4.5
Last updated on
Workshop: Build Your First RAG Pipeline