Vitrubo

Any medical document in. One interpreted record out.

Vitrubo reads lab PDFs, hospital records and photos into one record and returns a conclusion with the guidance it rests on. To the lab, the doctor or the person who decides.

Every value with its source
Patient and doctor views
The Health Report — every value with the guidance behind it
HL7FHIR R4LOINCSNOMED CTICD-10RxNormATCUCUM unitsEHAWHONICEKDIGOBSH

Documented everywhere. Understood nowhere.

Scattered across clinics

Clinics, labs, portals and drawers each hold a piece.

Written for clinicians

Written for clinicians but delivered to patients.

Locked in PDFs

PDFs, photos, paper. Unsearchable, unconnected.

Read without context

No context, no trend, no next step.

Read without memory

A value drifting for years looks normal in any single report.

Every source becomes one record.

PDFLab results
HL7Hospital record
JPGPrescription photo
APIPartner feed
Vitrubo
ParseStructureUnderstand
FHIR R4
Vitrubo
One living record
Health Record and Health Report with patient and doctor views

Built for everyone in the process.

A comment on the result. Drafted, not signed.

  • Written in the style the lab already uses.
  • The scientist or clinical consultant approves it in the existing release step.
  • The right next test is raised while the sample is still fresh.
For labs
Lab scientists reviewing a result together at the bench

Three capabilities, one platform.

We take the steps of medical work that need medical knowledge, and automate them.

Recognition

A document becomes data.

Works from a PDF, a photo, a scan or a live feed.

Normalization

A line becomes a reference entity.

LOINC, SNOMED, ICD, RxNorm, ATC. Units made comparable.

Interpretation

Medical knowledge applied to one person.

Returned with its basis to whoever decides. The main thing we build.

Who Vitrubo is built for.

LaboratoriesA draft comment at the release stepHospitals & clinicsThe whole history before the visitDoctorsRead the patient, not the folderHealth appsThe feature users come back for
TelehealthThe record arrives before the call
Insurers & employersOne reading across a population of tests
Pharmacy chainsResults explained, the next test suggested
Wellness & supplement partnersThe basis for a recommendation, from the person's own results

Everything, read together.

Labs, letters, discharge summaries and photos from different years and different labs are interpreted as one picture, against the person's conditions, medications and history.

Needs attentionneeds a doctor conversation now
Worth watchingoutside the range but not urgent
Back to normalwas flagged and has resolved
Always normalnothing to do
The Health Report on a tablet — flags grouped into four zones

Inside your perimeter & in your standards.

Your perimeter

Runs where your data already lives.

Containers on your infrastructure or a private cloud. You own the infrastructure and the keys. Vitrubo ships versioned builds and has no access to your data.

Identity never enters the prompt.

Models see values and an internal ID. Never a name, never personal data.

FHIR

Native, not an export.

One FHIR R4 record per person is the store itself. Every result coded and dated, the source document kept.

Speaks your standards.

HL7 and FHIR R4 in. FHIR and JSON out. Results mapped to reference codes with units made comparable.

HL7FHIR R4LOINCSNOMED CTICD-10RxNormATCJSON outFHIR outUCUM units

Same case. Two conversations.

The patient view of a report

Patient

Plain language. What it means and what to discuss with a doctor.

For patients
The doctor view of the same report

Doctor

Full data and clinical reasoning: connected conditions and medications and the tests to consider.

For doctors

A glass box, not a black box.

The model writes explanations. It computes nothing.

Multi-model interpretation

Several AI models cross-check each other: disagreements resolved, omissions caught, completeness enforced.

Numbers computed in code

Parsing, units, ranges and flags are computed. No number is ever written by the model.

Built on named guidance

EHA, WHO, NICE, KDIGO and BSH. One click opens the publication.

Identity never enters the prompt

Identity and clinical data are separated at the architecture level. Models receive de-identified values keyed to an internal ID; no PII or PHI is ever passed to a model.

One biomarker, its interpretation and the source behind it
EHAWHONICEKDIGOBSHFDA

Up to 30 sources behind a single biomarker view. Each chip opens the same guideline page a clinician would read.

Measured. Every release.

100%

Stanford MedAgentBench

600 of 600 tasks across five runs with zero failures. The first and only perfect score recorded.

100k

Hospital patients in an on-premises program

Clinical program with Assuta Medical Center. Real FHIR R4 data; nothing left the perimeter.

77%

of severe AI harms are omissions

ARISE State of Clinical AI Report 2026 by Stanford and Harvard. Our completeness gates run in code rather than in the model.

Demo cases

Invented people, real product. Every case is a synthetic record with a stock portrait: no real patient, no real result. The reports are live.

How Vitrubo fits into your process.

Place in the process

Visibility

Data and compute

Output form

Response time

Vitrubo hands over a draft for signature at report signature, invisible, in your stream, running in your perimeter, the same day.

Book a call on this configuration