Your Realtime AI is NGMI - Building Real-Time AI Systems from the Network Layer Up

Learn why most realtime AI applications fail and discover the correct approach to building reliable voice and video AI systems in this 17-minute conference talk. Explore the fundamental design principles that separate successful realtime AI implementations from those that struggle with latency and reliability issues. Understand why starting from the network layer up, rather than from the model or application layer down, is crucial for delivering consistent realtime audio and video streams. Examine the technical differences between WebRTC and WebSockets for real-time audio applications, including detailed latency breakdowns and performance comparisons. Discover the advantages of WebRTC for voice AI applications and see practical demonstrations of its implementation. Delve into critical realtime AI primitives including interruption handling, conversation state management, and asynchronous function calling. Watch a live demonstration of Squabbert and explore community projects showcasing innovative WebRTC implementations. Gain insights from two industry experts who agree on core realtime systems principles but engage in spirited debate about client-side SDK approaches and vendor-neutral API design. Access practical resources and learn about the flexibility of WebRTC connections for various voice AI use cases.