<aside>
💡 Our goal is an AI for voice conversations that's as natural as talking to a person.
MetaVoice is founded by
Today's voice AI fails at real-world conversations. It’s slow, turn-based like a walkie-talkie, breaks with interruptions, and doesn’t understand emotion.
Developers can't build compelling experiences and users disengage. This limits voice AI to simple receptionist tasks and basic customer support, blocking meaningful services (sales, therapy, coaching) where dialogue and emotional intelligence matter most. **Scaling current tech does not work.
Our approach is a duplex speech-to-speech model** that learns conversational behaviour directly from data
That’s how we make voice the most natural way to interact with AI.
</aside>
The role
This is a founding engineer role owning the ML data & evaluation platform that model development runs on.
Who we’re looking for
You’ve worked on distributed ML infra and understand that model quality is often a data problem. You:
- Have a track record of taking initiative and learning things quickly.
- Care deeply about the people using the systems you build, whether they're researchers, operators or customers
- Have built something yourself (startup, side project, etc) or worked in an early-stage startup
What you’ve done
- Built distributed ML data pipelines to generate datasets for training and evaluation, including curation, sampling and slicing.
- Built and operated MLOps infrastructure to train, evaluate and serve ML models, including versioning, orchestration and observability.
- Built eval systems to measure model quality and catch regressions.
- Developed internal tools and workflows for researchers and operators to explore data and iterate fast.
Strong plus:
- Designed data flywheels: automated data collection and model update loops.
- Designed evaluation frameworks: automated or human-in-the-loop
- Experience with speech, audio, or conversational datasets
What we offer
- Change the world (when we succeed)
- Environment to do the best work of your life.