Voice scheduling
The scheduling app you talk to.
Your hands are full at exactly the moment the schedule needs changing. So you say what changed, and the day re-plans itself in front of you, with nothing committed until you say yes.
Almost every scheduling app assumes you are sitting down. Home health, hospice, therapy, mobile phlebotomy and most field work are the opposite: the moment the schedule needs changing is exactly the moment your hands are full. So we built the scheduling app you talk to.
What a voice scheduling app has to get right
Talking to software is easy to demo and hard to trust. Three rules make the difference between a novelty and something a clinician will use on a Tuesday:
- Confirm before commit. The agent proposes; you approve. Nothing changes state on the strength of a sentence it thought it heard. This is enforced in code, not asked for in a prompt.
- Never act on the wrong patient. Two people answering to one name is a question, always, never a best guess.
- Say what it cannot do. An agent that quietly does something approximate is worse than one that asks.
What you can say
| You say | What happens |
|---|---|
| "Arthur cancelled" | The visit comes out, the day re-orders, drive times recompute, and the freed hour is offered to whoever is nearest. |
| "Move Beatriz to ten" | The move is proposed with the knock-on effects shown, and the agreement on that time is cleared, because it was agreed about the old slot. |
| "Who can take nine o'clock?" | Candidates ranked by the driving each would add, not by who looks free. |
| "New client, Rosa Ibarra, 2200 Shattuck, Tuesdays and Fridays" | The address is placed and the record is prefilled for you to confirm. |
| "How busy am I next month?" | Answered from the plan on screen. |
Voice, but not a smart speaker
The comparison people reach for is a home assistant, and it is the wrong one. A smart speaker answers questions. A scheduling agent changes a record that a real person is depending on, which raises the bar considerably:
- Speech recognition runs on the phone, so audio never leaves the device.
- Reading a referral photograph also happens on the phone; only the extracted text is sent.
- Every extracted value must appear on the source page or it is dropped. Nothing is invented to fill a field.
- Model calls go through a business-associate-covered endpoint, and no real patient data goes near a model before that agreement is in place.
Does it work while driving?
That is the intended case, and it is why the interface is voice at all. The agent listens, proposes out loud and on screen, and waits. Nothing is committed without a confirmation, so a misheard sentence costs you a "no", not a patient's appointment. Handing the drive itself to Apple Maps or Google Maps keeps the navigation you already trust.
Who is using it
It is in private testing with a small group of clinicians. The people it is built for are home health nurses and aides, physical, occupational and speech therapists, hospice teams, mobile phlebotomists, social workers, case managers, and field service workers whose day is a route.
Questions people actually ask
What is a voice scheduling app?
An app you change your schedule by speaking to rather than typing. You say what changed, it works out the consequences and proposes a new plan, and it waits for you to confirm before anything is saved.
Is it safe to use while driving?
It is designed for hands on the wheel: the agent listens, proposes and waits, and nothing is committed without a confirmation. Follow your local law on phone use, and let the app read the day to you rather than reading it yourself.
Does my voice get sent to a server?
No. Speech recognition runs on the phone. Only the resulting text leaves the device, and photographs of referrals are read on the phone too.
What if it mishears a patient's name?
Two people answering to one name is always a question, never a guess, and the agent proposes rather than acts, so a misheard sentence costs a confirmation rather than a wrong appointment.
How accurate is it?
On our own graded set of synthetic referrals and spoken commands: 97% field accuracy on extraction with the model, 91% on voice intents, and zero invented values. We rerun the grading after any change.
Which devices does it run on?
iPhone today. It is in private testing, with a public TestFlight next.
Obillions
A voice agent for a day of visits. You say what changed and the day re-plans itself: the order, the drive times, and who nearby could take the hour a cancellation just freed. Nothing changes until you confirm, and your voice and photos never leave your phone.
In private testing now. A public TestFlight comes next, then the App Store.