Ever wondered what it takes to power millions of voice conversations at ChatGPT's scale?
When OpenAI needed infrastructure for ChatGPT's Advanced Voice Mode, they turned to LiveKit's open source Agents framework. Not a proprietary black box. Not a closed platform. Open source software that anyone can use, modify, and deploy.
In this talk, I'll take you behind the scenes of building production voice AI infrastructure that handles millions of conversations:
- Why Open Source for Production AI – The technical and business reasons behind the choice
- Architecture Decisions – How we built for scale, reliability, and low latency
- Scaling to Millions of Calls – The challenges you don't anticipate until you hit them
- Lessons Learned – What we'd do differently knowing what we know now
- What's Possible Now – How you can use the same infrastructure for your projects
This isn't a sales pitch, it's a technical deep dive with real production metrics, architectural diagrams, and honest discussions about trade-offs. You'll see the actual stack, understand the scaling challenges, and learn from our mistakes.
Whether you're building your first voice agent or scaling to production, you'll walk away with insights from one of the largest voice AI deployments in the world. Because the infrastructure powering ChatGPT's voice mode is open source, and it's available to everyone.
This talk has been presented at JSNation 2026, check out the latest edition of this JavaScript Conference.

















