VapiCon 2025 Voice AI Infrastructure: How to Scale & Achieve 99.999%

Description

At VapiCon 2025, Tejas Siripurapu leads a panel with Sean DuBois (OpenAI), Erik Bernhardsson (Modal), Renan Dincer (Cloudflare), Dmitry Ilyevsky (Apoxy), and Zach Koch (Ultravox) exploring what it takes to build reliable, low-latency voice AI systems at scale. The panel breaks down the infrastructure stack from ASR to TTS, discusses why fewer moving parts improve latency, and explains how open protocols like WebRTC enable reliability. They share lessons from major outages, the benefits of multi-tenancy for smoothing demand, and why running your own inference fleet matters for consistent performance. 💡 Topics Covered: Voice AI infrastructure stack anatomy Achieving low latency through architecture and edge computing Open protocols for reliability and autonomy Open weight models vs. API providers Multi-tenancy and cloud scaling strategies Lessons from major incidents and outages 👥 Panelists: Tejas Siripurapu (Moderator) - Head of Infra, Vapi Sean DuBois - Member of Technical Staff, OpenAI Erik Bernhardsson - Founder & CEO, Modal Renan Dincer - Principal Systems Engineer, Cloudflare Dmitry Ilyevsky - Co-founder & Engineering Leader, Apoxy Dev Zach Koch - Founder & CEO, Ultravox 📅 Subscribe to our calendar to see upcoming events https://lu.ma/vapi_events 🌐 Our website https://vapi.ai