Medellis Scribe — an AI medical scribe clinicians can actually verify

01The problem

Clinical documentation is the most resented hour of medicine. Notes get written after hours, billing codes get chosen defensively — under-coded revenue or over-coded audit risk — and the first generation of AI scribes made a subtle mistake: they produced fluent prose that clinicians couldn't trace back to the encounter. When you can't see where a sentence came from, reviewing the note takes nearly as long as writing it. Trust, not fluency, is the bottleneck.

02What we built

• Ambient capture of the encounter that becomes a structured clinical note in the clinician's format — across 20+ specialties, in English and Spanish.
• Evidence-linked generation: every statement in the note traces to its source in the encounter, so review becomes verification instead of rewriting.
• HCC-aware ICD-10 and CPT suggestions, tuned for risk-adjusted billing rather than generic code lookup.
• Patient-friendly summaries and prior-auth letters generated from the same encounter data — the paperwork after the paperwork.
• iOS and web clients; audio is purged after processing; every action leaves an audit trail.

03How it's built

The pipeline is deliberately boring: verify ownership, transcribe, generate against evidence, code, then delete what shouldn't be kept. Quality lives in an evaluation harness, not in vibes — changes ship when the scores say they're safe.

04Where it stands

Live and in active development as our flagship product. It is also the reason our healthcare consulting is grounded: the HIPAA-grade habits — minimum data, purge-by-default, human-in-the-loop, audit everything — were earned here, and they travel to every engagement we take.