serious harms a year from diagnostic error in the US. · bmj q&s 2023
The note records what has already been decided.
Medical scribes document. Evidence tools answer what you ask. Sofya works earlier, while the decision can still change.
of recommended care actually gets delivered. · nejm 2003
to follow the guidelines. The visit has twelve minutes. · jgim 2022
An expert beside you, in real time.
Not just a note or an answer. The case evolves while you work.
Man in his 60s, hypertensive, off one of his pills for weeks. This morning his right arm went heavy, then recovered.
“Heaviness, went away on its own.” A resolved focal deficit is a transient event until proven otherwise. Imaging within 24h.
abcd² ≥4 pending onset · aha/asa 2023 · class i“One of the pills” matches three on record. Last fill dates point to lisinopril, stopped ~5 weeks ago and still unconfirmed.
rxnorm 314076 · fill history · 3 candidatesCreatinine 1.9 mg/dL sat in a lab PDF from April. The eGFR was 38 and never appeared in the problem list. It changes today’s imaging contrast.
loinc 2160-0 · 2026-04-12 · egfr 38the note, already drafted
sourced line by line, reviewed before you sign
the conversation, coded and flagged
accept, edit or dismiss, each one recorded
the chart, already read
prior labs, meds, what the last visit left open
Sofya works the whole visit. The note is what falls out of it.
Continuous care for every patient, including between visits.
Sofya reads what arrives, follows what remains open, and brings the right context to the next decision.
see a longitudinal case →- Labs and imaging: ordered, resulted and read the same day before the visit; acted on today.
- Care gap: retinal screening came due and was surfaced before the visit; booked today, screened and closed after.
- Patient wellbeing: six check-ins arrived and were flagged before the visit; summarized today and followed up after.
- Medication use: prescribed at a visit, reaction reported and still not resumed before today; rechallenge today and reaction checked after.
Built for the work that already happens.
From triage to continuity of care, without creating a parallel workflow.
intake and triage
The right queue and the right urgency, before a clinician spends a minute.
consultation
Nothing missed while the patient is still in the room.
referral and handover
The question and the workup travel together, reducing repeated tests.
continuity of care
The next visit opens where the last one stopped.
primary care
Four problems in twelve minutes, and the one that cannot wait is named.
specialty outpatient
The governing guideline, by name and date, at the moment of decision.
cancer centre
Staging, eligibility and toxicity checked against protocol every cycle.
emergency department
Time-critical patterns surfaced while the workup is still being ordered.
virtual care
The chart carries what the hands cannot examine.
clinical audit
Every suggestion carries its source, its state and its timestamp.
quality improvement
Concordance measured from coded reality, not a chart-review sample.
value-based care
Care gaps closed inside the visit, and coded well enough to count.
Whichever one you start from, the same four things run underneath it.
listens
The consultation, and everything the case already carries from before it.
structures
Every finding becomes a coded entity that other systems can read.
suggests
From published evidence, with its source and its strength.
documents
A reviewable note is drafted, coded and ready for the system of record.
Interoperability in both directions.
Inside the system you already use, without replacing the system of record.
from the chart to Sofya
- Problem list
- Medications
- Labs and vitals
- Prior notes
- Imaging reports
from Sofya back to the chart
- Coded entities
- Optimized care plan
- The signed note
latam
Security as the floor.
read the security overview →the operating boundary
Controls follow the case, not only the infrastructure.
Access is least-privilege, activity is auditable, and protected health information stays inside the contracted operating boundary. Responsibilities are defined before deployment through the BAA.
- SOC 2 Type II
- BAA-defined responsibilities
- auditable access
Institution-controlled deployment.
Run inside the institution's cloud and perimeter, isolated and fully auditable.
Model governance by default.
Sofya and approved external models operate under the same access, logging, and review controls.
the patient’s decision
Consent belongs inside the clinical workflow.
The patient sees what is captured and why before the session starts. The institution defines the language, records the decision, and can honor withdrawal without losing the care record.
- plain-language notice
- purpose and scope
- timestamped decision
Consent travels with the case.
The decision remains connected to its purpose, scope, and timestamp throughout the workflow.
Withdrawal changes future use.
Revocation can stop later capture and processing without erasing the required care record.
the data lifecycle
The institution sets the clock.
Retention windows, deletion, export, and legal holds follow institutional policy. A product default never replaces a clinical governance decision.
- institutional policy
- deletion and export
- legal-hold controls
Policy defines every window.
The institution sets retention, export, deletion, and legal-hold rules for each deployment.
Lifecycle events stay auditable.
Actions can be verified without retaining the clinical payload longer than authorized.
the learning boundary
Care data is not training inventory.
Identified production data is excluded from model training. Evaluation and improvement use only datasets and processes explicitly approved for that purpose, with provenance and review.
- no identified-data training
- approved evaluation sets
- provenance and review
Production care data stays out.
Identified clinical content never becomes general model-training material.
Evaluation uses approved evidence.
Every dataset has a defined purpose, documented provenance, and human review.
The reasoning, written down.
Cases by specialty and the evidence that supports each decision.
latin america · to date2,000,000+
clinical encounters with physician review and signature.
94%diagnostic accuracy · internal validation 202596%physician satisfaction · deployment survey 2025
bring these results to your institution →Published. Recognized. At scale.
talk about your institution →Published
Foundation model paper at AAAI 2024, supported by seven earlier studies.
Recognized
Selected for programs at Mayo Clinic Platform and Stanford AIMI, with a presentation at the Meta LlamaCon 2025 keynote.
At scale
In routine use across Latin America, Einstein among them.