We did not build an AI and then look for a problem. We started with 20,604 real patients in Uganda and built the AI they needed — transparent, explainable, and clinically grounded.
We did not use synthetic data. We did not use data from another country and hope it would transfer. We used 20,604 actual HIV-positive patient records from Uganda — because that is the only way to build AI that actually works here.
All patient data used in this research was fully anonymised before model training. No names, no identifiers, no personal health numbers — only anonymised clinical patterns. Our ethics approval was granted by the REC at Uganda National Health Laboratory Services, ensuring every step of our research meets the highest standards of patient privacy and research integrity.
Our model analyses these 15+ clinical factors simultaneously — cross-referencing relationships that would take a clinician hours to review manually:
Five steps. Plain English. No jargon. This is exactly how the prediction works — from the moment a patient's data enters the system to the moment a clinician receives an alert.
The patient's existing clinical data enters the system automatically from your health information system. Nothing new for the clinic to collect or enter. No extra workload.
Integration-ready — works with existing EMR systemsOur XGBoost machine learning model analyses 15+ clinical variables simultaneously — cross-referencing patterns that would take a human clinician hours to identify manually.
XGBoost · Trained on 20,604 Ugandan patientsThe model outputs a risk score: Low, Medium, or High — indicating the probability that this patient will experience HIV viral rebound before their next scheduled clinic visit.
95%+ accuracy on validation datasetThis is what makes Gritap AI different from every other tool. A SHAP waterfall chart is generated showing the top 3–5 clinical factors driving the prediction — colour-coded, ranked by importance, and written in clinical language the nurse already uses.
Red bars = risk factors · Blue bars = protective factorsThe clinician sees a clear flag: "This patient is HIGH RISK" — and immediately below it, the specific reasons why. She verifies, decides, and acts. The AI supports the decision. It never replaces the clinician's judgment.
Early action prevents hospitalisation and saves livesEarly identification. Better decisions. Lives saved.
Every other tool in this space is either manual, opaque, or built without Uganda's patients in mind. Gritap AI is the first to combine all three requirements: prediction, explainability, and local context.
| Feature | Traditional Healthcare Systems | ✦ Gritap AI |
|---|---|---|
| AI-powered predictions | ❌ No AI — manual review only | ✅ Real-time XGBoost predictions |
| Explainability | ❌ No explanation — clinicians guess | ✅ SHAP waterfall for every prediction |
| Local context | ❌ Generic tools not built for Uganda | ✅ Trained on 20,604 Ugandan patients |
| Automated flagging | ❌ Manual — misses patients in busy clinics | ✅ Automated high-risk patient alerts |
| System integration | ❌ Standalone tools, poor integration | ✅ Integrates with existing EMR systems |
| Ethics & privacy | ⚠️ Varies — often unclear | ✅ REC-approved · Fully anonymised |
| Cost to facilities | 💸 High — manual labour + generic software | ✅ $400–$1,000/year SaaS subscription |
Peer-reviewed research, conference presentations, and open-source contributions from the Gritap AI team.
We are actively seeking clinical partners, academic collaborators, NGO co-funders, and health facilities ready to pilot the tool. The dataset is real. The tool is live. The results are measurable.