Skip to content
MediLink AI
Built on clinically evaluated NLP models

Describe your symptoms.Reach the right specialist.

Speak or type how you feel — in English or ChiShona. MediLink AI analyses your description, identifies the most likely condition, and routes your case to a qualified specialist, with the reasoning shown in plain language.

2
Languages supported
23
Registered specialists
< 60s
Typical turnaround
Referral result
What the patient said

“Ndine fivha nemusoro unorwadza kwemasvondo maviri, uye handidye zvakanaka.”

Fever and headache for two weeks, with poor appetite.

Most likely condition

Malaria

84% confidence

Infectious Disease

3 specialists matched · Harare

Urgent

Illustrative example. MediLink AI supports a clinician's decision and never replaces a diagnosis.

How it works

Four steps from symptom to specialist

Every referral follows the same path, and each stage is visible to you — including how confident the system is and why.

  1. 01

    Describe how you feel

    Record a voice note or type a few sentences. No medical vocabulary needed — everyday words work.

  2. 02

    Understood in your language

    Speech is transcribed and ChiShona is interpreted alongside English, so nothing is lost in translation.

  3. 03

    Analysed, then verified

    A trained classifier proposes the most likely condition; a clinical reasoning step checks it and flags urgency.

  4. 04

    Routed to a specialist

    The case reaches specialists in the matching field with contact details and availability shown.

Built for Zimbabwe

Care should not depend on speaking English

Most symptom-checking tools assume English. MediLink AI accepts ChiShona directly, so patients describe their symptoms in the language they actually think in.

  • Voice notes in English

    Speech-to-text handles accents and colloquial phrasing, then shows you the transcript before anything is submitted.

  • Typed ChiShona

    Shona descriptions are interpreted clinically rather than word-for-word, and the original text is always kept alongside.

  • Local disease context

    Prevalence of malaria, typhoid, TB and hypertension in the region is weighed in, not treated as an afterthought.

EnglishChiShona

Ndiri kurwadziwa nemusoro uye ndine fivha.

I have a headache and a fever.

Ndine dumbu rinorwadza uye ndiri kuseva.

My stomach hurts and I have diarrhoea.

Ndiri kushaya mweya uye chipfuva chinorwadza.

I am short of breath and my chest hurts.

Evaluation

The models are measured, not assumed

Three classifiers are trained on a labelled symptom dataset and compared on held-out data. The best performer serves predictions, and these are its figures.

Classifier comparison on held-out evaluation data
ModelAccuracyPrecisionRecallF1
Naive BayesIn use100.0%100.0%100.0%100.0%
SVM (Linear)100.0%100.0%100.0%100.0%
Random Forest97.6%98.8%97.6%97.6%

Figures describe model performance on an evaluation dataset, not clinical outcomes. Every referral is reviewed by a qualified specialist before any treatment decision.

Privacy

Your words stay yours

Health information is the most sensitive data a person has. These are the specific guarantees the system is built to keep.

Recordings encrypted

Each voice note is sealed with its own key, and that key is itself encrypted — a database copy alone cannot open it.

Deleted on a schedule

Recordings are purged automatically once the retention window passes. The transcript your specialist needs remains.

Identity withheld

A specialist sees the clinical picture before they see who you are. Your contact details appear only once they accept.

Erasure on request

One action removes your account, referrals and recordings for good. Only a pseudonymous audit entry is kept.

Start with the symptoms you already have words for

Creating an account takes under a minute, and your first referral takes about as long as describing your symptoms out loud.