- Healthcare
- AI Agent Development
- AI Consulting
- Cybersecurity
Giving clinicians back the hour they spend writing notes
A multi-site practice loses a substantial share of every clinician's week to documentation, most of it typed up after the last patient has left. An ambient documentation agent drafts the note during the visit and leaves the clinician doing the one thing only they can do, which is checking it.
Representative engagement. This describes a pattern we build rather than one named client: the situation that produces it, how we approach it, and the range of outcomes that kind of work lands in. Figures are stated as ranges or targets, never as a measured result for a specific customer. Our named client work is on the work index.
- Industry
- Healthcare
- documentation time saved per clinic session (typical published range)
- 15 to 30 min
- of notes reviewed and signed by a clinician before filing
- 100%
The problem
Notes get written from memory at the end of the day, because writing them during a twelve minute consultation means looking at a screen instead of a patient. Written from memory, they are thinner and later, and everything downstream depends on them: coding, referrals, continuity of care, the audit trail. The temptation is to buy a transcription tool and call it solved. Transcription is the easy part. The hard parts are knowing which of the things said in the room belong in the record, getting the output into the EHR without a copy and paste step that nobody will do, and deciding what happens on the cases where the model is unsure. There is also a floor under all of it. Protected health information cannot be handled casually, consent has to be real and revocable, and a clinician has to sign the note. An agent that files anything unreviewed is not a product, it is a liability.
What we built
An ambient agent that listens with consent, drafts a structured note in the format the practice already uses, and presents it for review inside the clinician's existing workflow rather than in a separate tool. Structure comes first. The draft is built against the note template the practice writes today, so review is reading and correcting rather than reformatting. Where the model is unsure, it marks the section rather than guessing, and an unsure section is visibly unsure. Nothing files itself. The clinician signs, and the signature is the gate. Every draft, edit and signature is logged, encryption is in place in transit and at rest, and the whole thing runs under a Business Associate Agreement. Where the practice cannot let audio leave its environment, that constraint is designed for before anything is built rather than discovered during procurement.
What changed
Documentation moves from after the last patient to during each visit, which is worth more than the raw minutes saved because it is the difference between a note written from the room and one written from memory. Published figures for ambient documentation put the saving in the range below, and that is the range this pattern is scoped against. What varies most is specialty and template complexity, so the honest answer for any particular practice comes from a two week pilot rather than from a proposal.
Built with
- TypeScript
- Python
- Anthropic
- FHIR
- PostgreSQL