Orchestration Lab

6 Multi-Agent Orchestration Patterns

• 🔗 Sequential — agents run one after another in a fixed pipeline
• ⚡ Concurrent — agents run in parallel, then a ranker merges results
• 💬 Group Chat — manager picks who speaks each round; shared transcript
• 🔀 Handoff — agents hand off to each other peer-to-peer
• 📋 Magentic — planner creates steps; coordinator dispatches workers
• 🏛️ Hierarchical — manager calls sub-agents as tools via as_tool()

Data Sources (shown in diagram)

• 🗄️ Cosmos DB — real database — product catalog via vector search (embeddings + VECTORDISTANCE)
• 📋 Inventory (sim) — simulated — hardcoded stock levels in Python (no real DB)
• 👤 Loyalty (sim) — simulated — hardcoded customer discounts in Python (no real CRM)

Only Cosmos DB is a real Azure database. Inventory and loyalty data are simulated in-memory for demo purposes. All tools use the same underlying functions that the MCP server exposes in the main shopping assistant.