iClerk takes AI scribe to the next level — it pairs a mainstream LLM with your own clinical pearls (short, distilled lessons from frontline practice), personal triggers, and local hospital protocols.
Designed by an emergency clinician with 12+ years of frontline ED experience, built from inside the workflow it serves.
UK Emergency Departments (EDs) log 25M+ attendances per year; documentation is estimated to consume around 40% of ED physician time.
Unlike ambient scribes optimised for primary care, iClerk is ED-native and retrieval-grounded in the clinician's own practice.
Most AI scribes on the market stop at structuring a consultation into a note. iClerk goes further — it pairs the general medical knowledge of a mainstream LLM with a retrieval layer indexed on the clinician's own knowledge base: personal triggers, clinical pearls, and local hospital protocols. Hospitals can feed in their own guidelines the same way, so the system is driven by local practice and the user's accumulated experience alongside general medical knowledge.
Emergency medicine reasoning is worst-case-first. Chest pain is a life threat until proven otherwise, not a musculoskeletal niggle. iClerk's prompting and retrieval logic are tuned for that mindset.
The architecture is versatile enough to be adapted for other specialties and eventually other languages — but ED is its native environment.
The goal is not to add another system for clinicians to manage — it is to give back time and cognitive bandwidth. iClerk handles the structuring of the note while surfacing the clinician's own pearls and local protocols at the point of care, so the things you meant to remember reach the decision when it matters.
That makes it both a quality-of-life tool for teams working under pressure and a quiet safety layer — not by generating advice, but by ensuring your own stored knowledge is available when the case is in front of you.
Outputs are clinically useful but always require clinical verification. iClerk is positioned explicitly as a clinical aid, not a decision-making tool. That distinction is non-negotiable — clinically and medico-legally.
The moment a system generates output that reads like clinical guidance rather than structured notes, the liability picture changes. iClerk stays on the right side of that line.
iClerk routes generation through a mainstream LLM — so output quality depends on the quality of the model behind the hood. Stronger models produce stronger notes.
iClerk's retrieval layer — your pearls, your triggers, your local protocols — is injected into every generation so the model's output can reflect your practice. How fully that shaping lands still depends on the underlying model.
iClerk is built against the operational requirements clinicians actually ask about.
iClerk does not record an entire audio file first, then send it for transcription. It instead processes audio in real time by streaming small, encrypted chunks as they are captured — enabling immediate transcription, reducing the need to store full recordings, and supporting stronger data minimisation. It also means work already spoken is already captured: an interruption mid-session doesn't take the note with it, the way a record-then-send workflow can.
Data is encrypted at rest. Every disclaimer acceptance, every note save, every configuration change is logged.
A semantic search widget runs over your own pearls, triggers, and hospital protocols. Ask in plain language — matches are by meaning, not exact keyword. In practice it's closer to a conversation with your knowledge base than a search bar.
Sessions sync across your devices — dictate at the bedside, pick up at the workstation. Sync isn't guaranteed instant, but in practice the gap is short.
Autosave runs continuously in the background and flushes on tab close or device switch, so work survives reloads, network blips, and handoffs between bedside and workstation. Local storage is deliberately minimal — only the session state and in-flight dictation text needed for the active consultation are held on the device, in line with data minimisation principles.
iClerk's value to a clinician grows with use. Every pearl stored, every trigger written, every protocol indexed sharpens the retrieval layer's fit to that individual's practice. Over time the system holds a record of how a clinician thinks that no generic scribe can replicate — and that a clinician would be reluctant to leave behind.
The same effect holds at team level: a shared index of local guidelines, annotated by the people who use them, accumulating into an institutional memory that survives staff turnover.
Currently best in English. Other languages are supported for general dictation, but the underlying speech-to-text engine does not yet reliably recognise medical terminology outside English. Broader language coverage is on the roadmap.
iClerk is currently in demo phase. It has been extensively tested in mock scenarios and is awaiting formal clinical safety sign-off.
It has been built against the UK regulatory framework, with the clinical safety case pending CSO qualification.
We work with emergency clinicians and hospital teams looking to reduce documentation burden without losing clinical nuance. Get in touch to arrange a demo or discuss integration with local protocols.