The numbers
A study in two US health systems covered 1,430 clinicians after AI began to draft documentation from visit recordings. In one system, the share of doctors with burnout fell by 21.2 percentage points in 84 days; in the other, well-being rose by 30.7 percentage points in 60 days (JAMA Network Open, August 2025). In a survey of 99 clinicians at the University of Kansas clinic, 73% did less paperwork after their shift, and 47% agreed they could add one more visit (February 2025). The Moscow health department used AI to create 30 million pre-visit summaries of electronic health records in a year (July 2026). A US network of 190 hospitals, where shifts hand over about 60,000 times a day, is testing AI summaries in 5 hospitals; in one of them, nurses rated the accuracy at 86% (2025).
What a clinic in Russia can introduce
A pre-visit chart summary from data that is already structured, and a draft visit note that the doctor edits and signs: these systems work with the health record, and the clinic installs them inside its own network. Our part here is different: consent and contract templates that fill in data from the schedule. The diagnosis, orders and signature stay with the doctor. Under the criteria of Roszdravnadzor, the Russian medical device regulator, AI that interprets patient data and affects a doctor’s decision counts as a medical device.
Data
A health record is protected by medical confidentiality and is a special category of data: the model runs inside the clinic’s network or at a processor in Russia under contract. For hospital analytics, Moscow deployed a language model locally so that data does not leave the perimeter (2026). Our document agent does not read health records: it works with the schedule and templates.
How to start
A pilot on one type of document with no medical data, such as consents or contracts: 14 days, from $1,499. The metric: minutes per document and the share of documents with no manual input.