A.L.I.C.E. · Automated Linguistic Information Categorization Engine

Not an AI scribe. A transcription engine that never changes your words.

Most "AI scribes" write your note from scratch — and quietly invent names, numbers, and dosages along the way. Alice doesn't. It transcribes exactly what was said, preserves every fact, flags anything uncertain, and uses AI only at the very end to make the letter read cleanly — never to change it.

A transcription agent — not an AI author

Get your evenings back without handing your charting to a model that makes things up. Alice captures your encounter verbatim, arranges it into your note template, and tidies the formatting at the end. Your words, names, numbers, and medications come through exactly as dictated — and anything Alice isn't certain of is flagged for you, not guessed.

  • On-device speech-to-text, GPU-accelerated and processed locally
  • Structured clinical notes generated from the transcript, preserving names, numbers, and medications as dictated
  • AI is used only to format the finished letter — never to write, change, or “fill in” clinical content
  • Review-and-sign workflow with full edit control
  • Speaker diarization tracks doctor vs. patient speech
A parrot, not an author

Alice doesn't hallucinate.

Hallucination is what happens when an AI writes a clinical note from scratch — inventing a dose, a name, a finding that was never said. Alice never writes your note. As the Automated Linguistic Information Categorization Engine, it transcribes deterministically, preserves exactly what you said, and flags anything uncertain for you to confirm.

AI touches the letter only at the very end, to make it read cleanly — it never decides what your patient said, what they take, or how much. A single dropped medication or altered number isn't an acceptable error rate in medicine; it's the whole point of building Alice differently.

Scribe Lab

The safety check between dictation and your note

Before a single word of AI formatting happens, Scribe Lab reads the transcript and flags what matters — uncertain drug names, unusual medical terms, spelled names, and identity- and billing-critical facts — and surfaces them to you. Nothing is silently guessed, and no note leaves until you've signed.

  • Flags uncertain drug names and dosages — a dedicated medication-safety check, not a spellchecker
  • Catches unusual medical terminology and spelled names heard but not confidently transcribed
  • Verifies identity- and billing-critical facts — health-card numbers, version codes, dates of birth, referring physicians
  • Preserves what was said and flags what's uncertain — it never invents to fill a gap

What the difference looks like

The same dictation, two engines. These are simulated examples — representative of the error classes Scribe Lab is built to catch.

Simulated · illustrative · no patient data shown
🔴Dose inflated 10× — potential overdose

Leading scribe

metformin 5000 mg

AliceMR

metformin 500 mg

🔴Medication silently dropped

Leading scribe

ramipril, atorvastatin, [warfarin missing]

AliceMR

ramipril, atorvastatin, warfarin

🟠Drug name invented vs. preserved & flagged

Leading scribe

Simbraze (committed as fact)

AliceMR

Simbrinza (flagged for review)

Every one of these is recoverable in AliceMR because Scribe Lab preserved or flagged it. In a one-pass system, the dropped or invented version is what reaches the chart.

Not all errors are equal

A misspelling a clinician catches is a nuisance. A medication that silently vanished, or a number the machine invented, is a danger. Scribe Lab scores by harm and keeps the mistakes in the recoverable column.

🟢

AliceMR — fails safe

  • ✓ Preserves every name, number, and medication exactly as dictated
  • ✓ Flags anything uncertain for clinician review — never guesses silently
  • ✓ A dedicated medication-safety check corrects flagged drug names
  • ✓ Every note is reviewed and signed by the clinician
🔴

"Fast" cloud scribes — fail fatal

  • ✕ Medications silently dropped — missing is lost, and AI cannot recover it
  • ✕ Numbers and health-card codes altered or omitted — rejected claims, wrong patient
  • ✕ A best-guess drug name committed to the record as if it were fact
  • ✕ Per-transcription cloud-AI fees, and PHI sent to a third party

Buy cheap, die twice.

Intelligent Documentation

Real-time Transcription

Record directly in-browser. Transcription runs on local GPU hardware for speed and privacy — your audio never leaves your infrastructure, and no third-party AI ever sees it.

Faithful structuring

Alice arranges your dictation into Subjective, Objective, Assessment, and Plan — preserving every word, number, and dose. The formatting is tidied; the facts are never touched.

Template Library

Pre-built and custom templates for physicals, follow-ups, mental health, and more. Alice formats the output to your preferred layout — the structure adapts, your words don't.

Alice Scribe benchmark lab

Built from real clinical dictation, not canned demos

AliceMR was built and tuned against thousands of real-world clinical dictations, then reviewed by a Professor of Medicine and Head of Geriatrics (Internal Medicine) to make sure the context and wording are razor sharp. We benchmark our output against real-world transcriptions from the industry's leading providers — on identical inputs — with AliceMR processing locally for stronger PHI control.

AliceMR is also built for the charting workflow around the note. Standalone scribes capture a transcript, but they sit on top of whatever record you already use — so a practitioner on a small or disconnected EMR has to move and reconcile each note, one patient at a time. AliceMR is the record, so the note, the corrections, and the billing all live in one place.

Measured letter quality

ASR errors, missed spelled names, medical terminology misses, grammar, formatting, and manual correction minutes are captured for export.

Integrated correction loop

Demographics, referring-physician details, processing time, and how cleanly the result drops into the chart are scored alongside the final note.

Critical-fact fidelity

Internal testing · identical real clinical dictations · June 2026 · validation in progress

Provisional
AliceMR — critical facts preserved ~98%
Leading scribe — critical facts preserved ~57%
Clinic names + health-card codes kept intact (AliceMR) 100%

AliceMR

Preserve & flag

Every name, number, and medication kept as dictated; anything uncertain is flagged for the clinician, and corrections flow back into the chart instead of being retyped.

Comparator

Drop & reconcile

Output quality is only part of the cost when missing facts must be caught by hand and the chart still needs manual reconciliation.

How It Works

1

Hit Record

Start recording during your encounter with one click. Alice captures every word locally while you focus on your patient.

2

Alice Structures the Note

Alice transcribes the conversation verbatim and arranges it into your note template — preserving names, numbers, and dosages exactly as said.

3

Review & Sign

Review the note, make any edits you need, and sign off — you are always the author. The final letter must be verified before it is sent.

See Alice Scribe in action

Book a demo and see how faithful, deterministic documentation gives you your evenings back — without inventing a thing.

Book a Demo