Learn how to architect production-grade AI systems that combine event-driven serverless patterns with multi-agent orchestration. This talk walks through a real archaeological research platform processing terabyte-scale LiDAR terrain data, demonstrating critical design patterns: durable workflows for long-running AI operations, fan-out/fan-in for parallel agent execution, external event correlation for human-in-the-loop approval, and state management across distributed agent conversations. See how Microsoft's Agent Framework (Semantic Kernel + AutoGen) simulates realistic expert collaboration, archaeologists debating terrain features, environmental analysts cross-referencing satellite data, and historians validating findings through natural multi-turn dialogue.

What attendees will learn:

- Implementing Durable Functions patterns: fan-out/fan-in, external events, human approval gates, and long-polling for async container workloads
- Managing conversational state across multi-agent systems using Microsoft Agent Framework orchestration
- Designing event-driven pipelines that trigger on blob storage events and coordinate distributed AI processing
- Simulating domain-expert collaboration: building agents that maintain context, debate conclusions, and reach consensus like real archaeologists
- Combining DiskANN vector search (Cosmos DB) with graph relationships (Neo4j) for stateful knowledge retrieval