What we do
Voice AI that survives a real phone call
Most voice demos fall over the first time someone interrupts, mumbles a postcode, or asks something the script never covered. We build agents that take real calls on a real number. They answer, work out what the caller wants, do it, and pass the call to a person when they should.
What we build
Picking up, and working out why
The agent answers, works out who is calling and what they actually want, and deals with it. It has to hold up when someone talks over the top of it, changes their mind halfway through a sentence, or opens with "yeah look, it's about the thing from Tuesday".
Booking, moving, cancelling
The appointment is made, moved or cancelled during the call, and it lands in your calendar before the caller hangs up. Nothing sits in an inbox waiting for someone to process it on Monday. Scheduling, payments and CRM are systems we have wired into production platforms before.
Answers from your material
Opening hours, prices, what you do and do not take on, what to bring to a first appointment. The agent answers out of your own documents. When it gets something wrong, you fix the document rather than the prompt.
Handover that keeps the thread
When it should stop, it stops. It transfers to a person with the transcript and everything already established attached, so nobody starts again from the top. Out of hours it takes a message worth reading.
How it is built
There are two ways to build one. A cascade runs speech to text, then a language model, then speech back out, so every stage is separate and you can swap any of them. A single speech-to-speech model takes audio in and gives audio out. That is faster and holds more of the tone, but there is less to look inside. Which one suits depends on how tightly you need to control what gets said.
Latency decides whether it feels like a conversation. Past about a second of silence, a caller hears someone who is not listening. So we spend that budget on purpose. Where the model runs, how early it starts speaking, what it says while it is still working.
It also has to survive interruption. People talk over the top, change their minds mid-sentence, and save the important part for last. An agent that only works when the caller waits their turn is a demo, not a phone line.
What we bring to it
Conversational AI, already shipped
We built an AI avatar product for customer advice, where holding the conversation well was the whole job.
AI inside a health product
We built a cross-platform health app that makes personalised recommendations. Same sector as most of the people who ask us about voice.
Agents and knowledge bases
Our own platform manages large knowledge bases and runs agents against them across several regions. Voice is another channel onto that.
The unglamorous half
An agent is only useful if it can reach your systems. We have put scheduling, payments and CRM into production platforms, including a HIPAA-compliant one used by doctors and patients, and a health insurer's customer portal. That engineering came before our AI work. It is why consent, recordings and personal data get handled properly here.
Where we work
We are in Byron Bay and work with businesses across the Northern Rivers, the Gold Coast and Brisbane. Health and allied-health practices are where we have done the most. Three past builds were healthcare platforms, one of them an AI health app. Bookings, cancellations and privacy obligations are things we have already had to get right.
Let's talk
Tell us what your phone does now, and what you wish it did.
Send a message