.svg){kind=icon}
From Real-Time Voice to AI Avatars: Live Play Mobile on the Next Era of Gaming
Loading video...
Show Notes
What does it take to evolve from live voice experiences to AI-powered game hosts? In this episode of the Convo AI World Podcast, Agora's John Moore sits down with Geremie Camara, co-founder of Live Play Mobile, to explore how conversational AI is reshaping live gaming and interactive entertainment. With more than 100,000 hours of live-hosted gameplay, Live Play Mobile has spent years learning what keeps players engaged. Now, the team is applying those lessons to AI avatars that can entertain, personalize experiences, and scale globally alongside human hosts. Geremie shares why real-time infrastructure matters, how AI and human hosts complement each other, what it takes to make AI feel natural, and the engineering challenges behind voice quality, latency, memory, and production-ready conversational AI. Whether you're building voice AI, AI agents, gaming experiences, or real-time applications, this conversation offers practical insights into where conversational AI is heading, and what it takes to build engaging AI experiences at scale.
Key Topics Covered
- •Live hosts to AI avatars (how AI complements human hosts rather than replacing them)
- •Human connection and player engagement
- •Building AI hosts for real-time entertainment
- •Real-time infrastructure and low-latency networking
- •The conversational AI technology stack (LLMs, voice, avatars, memory, orchestration)
- •Humans vs. AI: strengths, tradeoffs, and collaboration
- •Voice quality and expressiveness
- •Deploying AI avatars in production
- •Personalization, memory, and guardrails
- •The future of AI-powered gaming and interactive entertainment
Episode Chapters & Transcript
Teaser
Geremie on 100,000 hours of live entertainment, Agora as the heartbeat, humans and AI in a recursive loop, and racing toward the Turing test.
Welcome
John Moore introduces Geremie Camara, co-founder of Live Play Mobile, and the team's work at the intersection of live gaming, AI avatars, and entertainment.
Geremie Camara and the Live Play story
Geremie traces Live Play from 2019, 24/7 live bingo since COVID, nearly 100,000 hours on Agora, and a background at Paramount and Sony game shows.
What live hosts add that games cannot
Games deliver mastery and accomplishment; live hosts add being seen and celebrated—something players felt acutely during COVID isolation.
They come for bingo, stay for connection
Players return because authentic host relationships form over time: cheering luck, consoling losses, and making people feel appreciated.
Distilling hosting craft into AI avatars
With 100,000 hours of data and game-show pedigree, Live Play knows what great hosting is—and is teaching AI avatars the overlapping emotional work humans still uniquely do.
HQ Trivia, casinos, and why live hosts work
HQ Trivia proved live-hosted games on mobile; casino dealers proved a sustainable model. Broadcasting plus casino gaming became Live Play.
What great hosts actually do
Like TV game-show hosts, they direct focus: welcome newcomers, teach, make veterans feel special, entertain, and retain—skills taught to humans and AIs alike.
Why glass-to-glass subsecond latency matters
From Ripcord TV to Live Play, Agora's worldwide ultra-low-latency video was non-negotiable for fast-paced games where people win in real time.
When AI avatars became possible
Late 2025: LLMs, voice models, avatars, and Agora ConvoAI aligned. Voice-only experiments in November, video in March. Anatomy: heartbeat, brain, memory, voice.
Building a production-ready avatar stack
Agora as a Mr. Potato Head template—integrated video, voice, and LLM partners plus a 24/7 heartbeat you pin the whole experience on.
Humans vs. AI: a pyramid of content
AI learns fast, hosts 24/7, and speaks any language in minutes. Humans have lives and fandom. Pyramid: AI at scale, elite human hosts, celebrity events on top.
Magic moments start with a name
Hearing your name, Cheers-style belonging, and remembering birthdays and hard times—the human craft of making players feel uniquely seen.
AI as a host's sidekick
Personalization without creepiness: AI can nudge human hosts like a good partner who remembers it was John's birthday yesterday.
What an AI host must do in real time
Recognize every player, teach the game, read live KPIs, chat, train on voice and body language, and learn recursively from millions of in-game data points per hour.
Voice fidelity vs. real-time latency
Cloning a six-year host's voice fooled them eyes-closed; production still needs custom low-latency expressivity. They're racing toward the Turing test.
Fine-tuning for authentic high energy
Models didn't ship the right expressivity out of the box—so Live Play fine-tunes a narrow aperture: authentic, high energy, and kind.
A modular stack you can go deep on
Flip-a-switch model experiments on Agora's heartbeat, plus bolting proprietary expressivity back in where the platform hadn't caught up yet.
From MVP to production, fast
John on ten years of real-time WebRTC plus agentic development—Live Play shipped production avatars faster than most teams Agora has seen.
Early production results beat expectations
Voice MVP in November, video by March, avatars in games by April/May. Against a no-host baseline, avatars closed more of the hosted-KPI gap than expected.
"Are you real?" and hologram Haley
They don't pretend AI is human—yet players still ask. Would fans play with a hologram of a beloved host who only works four hours a day?
Guardrails, memory, and proprietary scaffolding
Game knowledge, player memory, third rails, and call-to-action craft stay in-house. ConvoAI keeps the coordinated utterance-to-avatar heartbeat on autopilot.
Frontier models vs. what you keep in-house
Benefit from interchangeable bleeding-edge models, but protect PII, domain expertise, and competitive IP with architects who build for performance and protection.
Don't reinvent the heartbeat
Focus on the closed loop of perceive-plan-perform-observe against real commerce data. Don't roll your own video, voice, or real-time transport.
Pipes that hit a million phones in 600 ms
Agentic coding makes rolling the whole stack tempting, but human or AI broadcasts still need Agora's sub-second WebRTC distribution.
One-to-many, then one-to-one
HQ Trivia's "special among the crowd" dynamic; seamless cuts between one-to-many and one-to-one; productizing the stack for live shopping and fitness in 2027.
Crossing the Turing test
Voice and avatar models feel ~80% of the way there—maybe 2027 or early 2028. Cloud sets, mini-games, and multi-avatar shows humans can't iterate that fast.
Celebrity avatars and AI co-hosts
Famous hosts won't sit in studio four hours a day. After the Turing test, likeness-based avatars and human-AI talk shows stop sounding silly.
The clone handoff experiment
Geremie's most exciting test: a beloved human host saying goodnight and leaving a clone that will brief them on everything the audience said overnight.
The future of gaming, then wrap
Unlimited experiments, solo-founder unicorns, and a rate of change no one contemplated. John thanks Geremie and closes the episode.
Click on any chapter to view its transcript content • Download full transcript
Convo AI Newsletter
Subscribe to stay up to date on what's happening in conversational and voice AI.