Mission
Voice AI for moments that can't wait.
In a courtroom, someone has to capture every word while a witness tells their story. In an emergency, a firefighter may need a quick answer while describing the layout of a building. In moments like these, a voice assistant that is slow, or that keeps talking when you cut in, is not just annoying. It gets in the way.
We are building a voice engine for these situations. It listens, answers quickly, and stops the moment you interrupt.
Use cases
Three moments we design for.
- Legal01
A witness tells their story in court. The agent keeps accurate notes while they speak, the way a court stenographer would.
- Operations02
A firefighter's earpiece takes in a building's layout. When the firefighter asks a follow-up question, it answers right away, and stops the moment they speak.
- Healthcare03
A patient arrives and the care team starts treatment. When the doctor asks about the patient's history, the agent answers at once from the connected records. In the operating room, a spoken question about blood pressure gets a short spoken answer from the connected monitor.
These are the situations we are designing for, not deployments today. In clinical settings, real use would need proper validation and the right regulatory and compliance steps.
Why interruptions matter
A machine that talks over you stops feeling like a tool.
When a voice agent keeps speaking after a person cuts in, people notice immediately. The conversation stops feeling natural, and the system starts to feel like something that isn't listening. We believe good turn-taking is what separates a voice tool people tolerate from one they trust. That is why our first benchmark focuses on how fast the agent falls silent when interrupted.
See the benchmarkWhat we are building
A modular voice engine, delivered through an API, that software vendors in legal, healthcare and operations can embed in their own products. Speech recognition, the language model and voice synthesis are separate parts, and the whole engine can run in the customer's own environment.
- 01
Modular, not a black box
- 02
Honest about the numbers
- 03
Runs in your environment
Where we are today
We are early. Our first internal benchmark is public, including what it does not yet measure: real call recordings, noisy audio, human listeners and compliance certification. We are looking for design partners to help decide what we validate next.
Team
Aysima Çelik
Founder
Om Chimurkar
Research Advisor