About iClerk®

The technology is coming. The standard is the question.

Clinical medicine will absorb this generation of AI — that much is already settled, in every specialty, with or without a plan. What is not settled is whether it arrives with the discipline the work demands: a named source behind every statement, a safety net that does not depend on the model's judgement, and a record of what was shown and why. iClerk is built for accuracy and safety first, and for everything else afterwards.

The note is the part everyone talks about. It is the smallest part of the problem.

Ask for access See what it does
The gap

A capable model is not a clinical tool.

Three things are missing the moment a general-purpose model meets a real clinical question.

Whether clinicians use this technology is not the interesting question. They already are — in browser tabs, with their own prompts, their own sources and no record of either. The real question is whether that happens once, inside a framework that can be governed and audited, or several hundred times in private, where no one can see what was asked, what came back, or what it rested on.

Which matters most the first time a fluent, confident, well-formatted and entirely unsourced answer turns out to be wrong — and nobody can tell.

01
It is not grounded
A general model answers from whatever it read on the open internet. Your department has its own guidelines, its own thresholds and its own referral routes, and they are the ones you will be held to.
02
It cannot be checked
An answer with no source cannot be verified in the seconds you have, and an answer that cannot be verified cannot safely be used. Fluency is not evidence.
03
It has no safety net
Asking a model to police itself puts the safety layer inside the thing it is meant to be checking. The catch has to sit outside the model, or it is not a catch.
The approach

Accuracy and safety are the design, not a feature.

The interesting work is not the model. It is the structure built around it so that a wrong answer is catchable, and a right one is checkable.

01
Grounded retrieval
Clinical content is drawn from your department's own published documents, versioned, with the current one marked. Not the open internet.
02
Screening outside the model
The diagnoses you cannot afford to miss are screened for by a separate layer that does not rely on the language model having been careful.
03
Provenance by default
Every clinical statement carries the source it came from. Citations are part of the record and cannot be edited away.
04
Knowledge that compounds
The judgements your seniors carry and never write down can be captured once and surfaced on the right patient, in the words they were written in.
05
Configurable to any specialty
The guidelines, the pearls and the must-not-miss list are written by the department. Change the sources and you change the specialty.
06
And it writes the note
Structured documentation falls out of the same pass. It is one output of six, not the product.
Where it stands

What iClerk is not.

It is not a certified medical device, and it does not hold formal clinical-safety or data-protection certification. It is in pre-deployment and is not in use with patients.

It is assistive. It does not make decisions, it does not carry responsibility, and it is not a substitute for the clinician reading the source it shows them. Clinical responsibility for every decision remains, entirely, with the treating clinician.

A tool that tells you where its answer came from can be checked in seconds. A tool that does not, cannot be checked at all — and in medicine that difference is the whole product.
Provenance

Where it was shaped.

The platform took its shape in acute hospital medicine, including the ER, because that is where the failure modes are sharpest: incomplete histories, decisions made under time pressure, and a set of diagnoses that must never be missed. A design that survives those conditions tends to hold elsewhere.

It is not built for one department or one way of working. Because the sources, the pearls and the must-not-miss list are authored locally, the same platform configures to any specialty that writes down how it practises.

Access

Come and break it.

Access is opened case by case while the platform is in pre-deployment. If you want to see whether it holds up against your own guidelines, get in touch and say what you would be testing it on.

Ask for access
Pre-deployment · not for use with patients