Careers
Build the speech intelligence layer.
The data that would make speech models work for everyone does not exist yet. We are the ones making it.
The frontier labs train voice models on clean speech, because their goal is transcription. So the models get very good at understanding people who were never hard to understand, and labelled disordered speech barely exists as training data. Everyone who blocks, repeats, prolongs, slurs or searches for a word is outside the distribution.
Sophia is where we start. Every consented session with her, reviewed by a speech-language pathologist, becomes the labelled data that makes the next model better — which lets one clinician reach more people, which produces more data. That loop is the company. Stuttering is the wedge, not the identity: the architecture already reaches speech sound, motor speech, voice, resonance, language and AAC.
Who we want to hear from
- Engineers who want to work on speech models and the product wrapped around them — training and evaluation, on-device inference, or the clinical tooling that turns a session into ground truth.
- Speech-language pathologists who think the current tools are beneath the work, and who want to shape how a personality behaves rather than fill in another form.
- Researchers who have spent real time with disordered speech and know why the benchmarks lie.
How we hire
We are small, and we hire ahead of roles. There is no listings page to check, because the honest answer is that a strong person in any of the three groups above changes what we can build next quarter.
Write to us. Tell us what you have built, what you would want to build here, and what you think everyone else has got wrong. We read every one.