Sydney Harbour
AI Product Operator
Ahson Kamal
AI product operator and technical founder. I design, build, and ship AI systems for complex, high-trust workflows.
Deep healthcare experience, now applied across industries where the work is operational, regulated, and unforgiving.
Where I’ve built
Healthcare is where I learned to build AI for trust, edge cases, and operational complexity. I now apply that same discipline to AI systems across complex, high-trust workflows.
Australia's first AI receptionist for GP clinics. I built it to answer calls, book appointments, and handle routine requests so front desks could stop drowning in the phone queue, and it went live in real practices with real patients, not a pilot.
clinics live · calls handled · booking rate
Founding PM for a voice receptionist that rolled out across 140+ radiology clinics. I owned it from the first messy calls to a containment-focused booking agent that held up against accents, interruptions, and the real way people book scans.
containment rate · monthly call volume
What I'm building now: AI voice and SMS for allied health clinics, psychology, OT, and physio. Missed-call text-back plus a receptionist that books, reschedules, and follows up, so a two-room practice gets front-desk coverage it could never afford to staff.
clinics live · early results
A marketplace matching GP locums to clinics, with the Accelevate receptionist stack underneath so cover requests and bookings handle themselves. Co-founded, currently in early sales.
stage · traction
AI systems for the messy middle, from strategy through to production. Healthcare is the strongest example, not the boundary.
Phone and SMS systems that answer, book, reschedule, and hand off to a human cleanly.
Operational flows that absorb edge cases and interruptions instead of breaking on them.
Admin and reviewer tooling so teams can see, trust, and steer what the AI is doing.
Measuring quality honestly, with call review and version tracking, before trusting a model with more.
Deciding what to build, what to automate, and when, with a clear path from prototype to production.
Taking AI from a working demo to a system that survives real, high-trust operations.
Build for the model that’s coming, not the one in front of you.
Capabilities move faster than roadmaps. I keep the hard parts in the data, the workflow, and the trust, with model-agnostic architecture underneath, so the model can be swapped for a better one without a rewrite.
Earn trust before you automate.
In high-trust workflows, a confident wrong answer is worse than no answer. I ship assistance first, track prompts and versions, review real calls and outputs, and only hand over control once evaluation says it has earned the seat.
Ship into the mess, not the demo.
Real operations have interruptions, accents, missing fields, handoffs, and edge cases that never appear in a polished demo. I design for admin, clinician, customer, and internal-user workflows from day one, because that is the only place the work actually has to hold.
I’m building Accelevate: evaluation-first AI agents and workflow systems. The work is voice AI, operational automation, and tooling for AI that can be reviewed, trusted, and improved over time.
Healthcare is the initial wedge, because it is complex, regulated, and operationally unforgiving. The same systems generalise to any workflow that has to hold under real conditions.
I’m an AI product operator and technical founder in Sydney, around ten years across healthcare technology and production AI, with a biomedical engineering background. I built Helen at HealthEngine, was founding PM at Avoca, and now run Accelevate. I care about AI that survives contact with real operations, not AI that looks good in a demo.
When I’m not building, I’m usually in the water or under the sky, freediving and shooting the ocean and the stars around Sydney. That’s roughly why this site sinks the way it does.
Cuttlefish, Cabbage Tree Bay
The water runs out, and the sky keeps going.
Milky Way over [location]
If you are building an AI product, automating a complex workflow, or trying to get AI from demo to production, I would like to hear about it.
Currently building Accelevate, evaluation-first voice AI and workflow automation, and writing about building AI systems that can be trusted in production.
