Biography
Roshan Manjaly leads forward deployed engineering at Vapi, the voice AI platform used by companies from YC startups to the Fortune 500 to put production voice agents in front of real callers. He holds a Master's in Artificial Intelligence from Georgia Tech and worked as a principal software engineer at Capital One before founding two startups of his own.
His work sits at the point where a voice agent stops being a demo and becomes infrastructure: mapping a customer's real workflows, handling latency and interruption, making tool calls reliable enough to trust in production, and turning each deployment into something the next one can reuse. Having built companies himself, he treats enterprise deployment as a product problem rather than a support one.
His perspective is especially relevant for audiences interested in applied AI, voice agents in production, and the engineering gap between a working prototype and a system that survives real users.
Session
- AI For TechSpeaker
Workshop 4: Give Your Product a Voice — Build a Live AI Voice Agent in 60 Minutes
Tuesday 22 September, 11:30 – 12:50Workshop Room BEveryone's talking about voice AI. In this room, you'll actually ship one. Roshan and Barbara will get you from a blank screen to a working, talking AI agent that can hold a real phone conversation, before the coffee break ends. You'll design its personality, wire up its brain (LLM), give it ears and a mouth (speech-to-text and text-to-speech), and connect it to a real tool so it does more than chat, it acts. Bring a laptop and an idea. Leave with a deployed agent, a live phone number, and the know-how to build the next one yourself. What you'll walk away with: • A working voice agent you built yourself, live • A mental model for the voice AI stack (STT → LLM → TTS → tools) and where the real engineering challenges hide (latency, interruption handling, tool-calling reliability) • The gap between a slick demo and something that survives a real caller — the practical lessons Vapi's forward-deployed engineers see across production deployments Best for: builders, PMs, and founders who want to stop reading about voice AI and start shipping it. No ML background required — bring curiosity and a laptop.
With Barbara Stanelli