Mednovo

Medical AI · Tashkent

Artificial intelligence for clinics that value the doctor's time and the patient's data

Mednovo builds products and research at the intersection of medicine and machine learning: from an AI assistant for doctors to disease screening from medical images. The clinic decides where its data lives.

Request a MedScribe pilot Become a partner

Directions

One team, two branches

Commercial products help clinics today. Research prepares what will enter practice tomorrow.

Mednovo

01Commercial

Products for clinics

Deployment, support, and a measurable result in the doctor's work.

  1. MedScribePilots in clinics

    AI assistant for doctors: visit recording, speech recognition, structured note.

  2. Custom clinic platformsOn request

    AI systems built around the workflows of a specific clinic or network: from image analysis to document handling.

  3. Integration and supportPart of deployment

    Connection to the hospital information system, staff training, ongoing support.

More about MedScribe

02Research

Science and early diagnosis

Grants, publications, joint projects with clinics and universities.

  1. Retina AIOngoing

    Diabetic retinopathy screening from fundus images.

  2. Skin lesionsCompleted

    Classification from dermoscopic images, mobile app.

  3. Brain tumorsCompleted

    3D glioma segmentation on multimodal MRI.

Research projects

01Commercial direction · flagship product

MedScribe: the visit is recorded, the note writes itself

The doctor talks to the patient, MedScribe listens. Uzbek and Russian speech becomes text, and the text becomes a structured encounter note. Manual input shrinks to reviewing and signing.

  • Minimal manual inputThe doctor does not type during the visit. The draft note is ready by the time the patient leaves the room.
  • More time for the patientThe doctor's attention stays on the patient, not the keyboard. Shorter visits, fuller notes.
  • Nothing gets lostComplaints, history, examination and plan are captured from the conversation, not reconstructed from memory.

Open medscribe.mednovo.uz

Recording 00:00

The visit goes on as usual. MedScribe records the conversation between doctor and patient.

Speech recognized and split by speaker

  1. Doctor
  2. Patient
  3. Doctor
  4. Patient
  5. Doctor

Structured encounter note

Complaints
History
Examination
Plan

Draft ready. The doctor reviews and signs.

How it works

  1. 1
    Record

    The visit goes on as usual while MedScribe records the conversation.

  2. 2
    Transcribe

    Uzbek and Russian speech is turned into text by a model fine-tuned on medical vocabulary.

  3. 3
    Structured note

    A local language model drafts the note: complaints, history, examination, plan. The doctor edits and confirms.

Where your data lives is your choice

On the clinic's server

MedScribe runs entirely on the clinic's own server. No internet connection is needed, and neither audio nor text leaves the building. The clinic decides who can access the data.

Hosted

Data is kept on the cloud provider's secure server and encrypted in transit and at rest. Audio is processed and deleted; only the text of the note is retained.

Free pilot

A free 4-week pilot for specialty clinics in Tashkent. Includes installation, adaptation to the clinic's specialty, and support. The hosted option needs no hardware; the on-premise option needs one server with an NVIDIA GPU.

Request a pilot

02Research direction

Computer vision for early diagnosis

Three projects at different stages: two completed, one ongoing with support from a state grant.

Ongoing · grant

Diabetic retinopathy

Retina AI: lesion segmentation on fundus images and retinopathy grading. The doctor sees each mask separately with adjustable opacity, plus the model's attention map.

  • Five lesion classes: exudates, hemorrhages, microaneurysms, laser scars
  • Attention map (Grad-CAM) to explain the decision
  • Working prototype, clinical evaluation in progress
Open retina.mednovo.uz
Lesion masks with adjustable opacity
Lesion masks with adjustable opacity
Model attention map
Model attention map
Completed

Skin lesion classification

Mobile app: a lesion photo from the phone camera or a dermatoscope, probabilities across seven classes including melanoma, and a highlight of the regions the model relied on.

  • Seven classes from dermoscopic images
  • Runs on the phone, result in seconds
  • Localization of suspicious regions
Mobile app: class probabilities
Mobile app: class probabilities
Model attention regions
Model attention regions
Original dermoscopic image
Original dermoscopic image
Completed

Brain tumor segmentation

3D U-Net for glioma segmentation on multimodal MRI: T1, T1ce, T2 and FLAIR. The model delineates the tumor core, edema and enhancing tumor across the whole volume.

  • Four MRI modalities in one volume
  • Three tumor sub-regions
  • Desktop viewer: prediction versus ground truth, slice by slice
3D U-Net: prediction and ground truth across four modalities
3D U-Net: prediction and ground truth across four modalities

Open to new problems in medical imaging and clinical text. Discuss a project

Partnerships

We are open to partnerships

Our products are built together with doctors. We are looking for clinics, researchers, and technology partners who share the same principles.

Clinics

MedScribe pilot deployments and joint development of solutions for your workflows.

Researchers and universities

Joint projects on medical imaging, publications, grants.

Technology and industry partners

Integration with hospital information systems, infrastructure, distribution.

Discuss a partnership

Contact

Write on Telegram. We reply within a business day.

Message on Telegram @sultanovshoh