Join the lab

We are a small team, and we intend to stay small. We hire rarely and carefully, for people who want to work on one hard problem until it is solved.

Where we are looking

These are areas, not job titles. If your work sits between two of them, that is a good sign.

Turn-taking and dialogue

You have worked on end-of-turn detection, barge-in, backchannels or multi-party dialogue, and you have opinions about how to measure them.
Full-duplex speech models

You have trained or adapted speech-to-speech models that listen while they talk, and you know where they break.
Reinforcement learning environments

You have built environments, simulators or reward models, and you care whether what a policy learns in simulation holds up in the world.
Evaluation

You think a benchmark is a research contribution. You have built one, or you have found the flaw in one that everyone else trusted.
Real-time systems

You make models fast enough for live conversation, where 200 milliseconds is a long time.

A problem that the field has not solved, and live conversations on Zoom, Google Meet and Microsoft Teams in which to test your answers.

Small teams, short loops, frozen evaluations. We measure progress against people in real rooms, not against our own previous numbers.

Send us something you made: a paper, a repository, a demo or a write-up. Tell us in a few lines what you would do first. We read every message.