The Solution — How Aafiya Works

Layer 0 — Safety First

AI Pre-consultation Triage

Before a patient completes any clinical intake, Aafiya runs a two-stage safety screen in under 500 milliseconds. The goal is simple: route emergencies to emergency services immediately, not after a queue wait.

The triage engine produces one of four outcomes. Patients can skip triage at any time — and every triage result is stored with model version and latency for the SAHPRA regulatory evidence file.

SAHPRA note: Every triage decision is logged with model version, latency, and outcome. Aafiya is building its SAHPRA regulatory dossier with every single consultation — not as a post-hoc exercise.

🚨 EMERGENCY

Chest pain, stroke symptoms, severe respiratory distress, haemorrhage, obstetric emergency. Immediate tap-to-call 112/10177 — bypasses clinical intake entirely.

⚠ URGENT

Priority queue placement, on-call physician alerted within 5 minutes. Paediatric, psychiatric, and high-risk cases surface here.

📋 ROUTINE

Standard queue. Clinical intake begins. Physician reviews within normal SLA windows.

🏠 SELF_CARE

Home care advice with red-flag escalation triggers provided. Patient knows exactly when to escalate. Clinical intake available if needed.

⚡

Stage 1: Deterministic Rules Engine

13 regex pattern groups covering cardiovascular, neurological, respiratory, haemorrhagic, obstetric, paediatric, psychiatric, trauma, and sepsis signals. Executes in under 50ms. No LLM involved — pure deterministic EMERGENCY detection.

🧠

Stage 2: LLM Classifier

If Stage 1 does not fire EMERGENCY, the LLM classifier runs with a structured JSON output schema: triage category, confidence score, reasoning, self-care advice where applicable, and red-flag list. 450ms hard timeout — falls back to ROUTINE if exceeded.

The Three-Layer Model

AI Intake. Physician Review. Patient cript Delivery.

Every consultation follows the same three steps. No AI output ever reaches a patient without physician sign-off. This is non-negotiable — by design and by regulation.

📱

Patient

Any phone
11 languages
PWA

→

🤖

AI Intake

SA-trained
clinical AI
EML & STG

→

👔

Physician

HPCSA review
3–8 minutes
Sign & approve

→

💊

Delivery

Patient
Signed note
E-prescription

1

AI-Powered Clinical Intake

The patient begins a structured clinical assessment via the Aafiya Progressive Web App. The AI asks clinically validated questions — symptoms, duration, severity, medical history, current medications — in the patient's chosen language from all 11 official SA languages.

The AI engine is fine-tuned on South Africa's specific disease burden: HIV/TB co-infection, NIMART protocols, the Essential Medicines List (EML), and Standard Treatment Guidelines (STGs). It does not guess; it follows structured clinical protocols. Red-flag conditions trigger immediate escalation to emergency services (10177) without waiting for physician review.

Intake includes embedded PHQ-9 and GAD-7 mental health screening for depression and anxiety — integrated into the standard clinical flow, not a separate referral step. South Africa has the 3rd-highest suicide rate in Africa; mental health conditions are systematically under-screened in primary care.

2

Mandatory HPCSA Physician Review

The AI output — a structured clinical summary with provisional diagnosis, ICD-10 code, and proposed treatment plan — enters a queue reviewed by an HPCSA-registered physician. The physician's dashboard presents everything needed to assess, modify, or reject the AI output.

The physician can: approve the AI output as-is, modify the diagnosis or prescription (every modification is logged as high-value training data), escalate for a live consultation, or reject and write a fresh assessment. The average review time is 3–8 minutes for straightforward cases, up to 15 minutes for complex ones. This enables 15–20 cases per physician per hour — versus the traditional 4–6 in-person consultations.

3

Patient Script Delivery

Once the physician signs off, the patient receives a signed consultation note and, where appropriate, an e-prescription routed to their nearest participating pharmacy (Dis-Chem, Clicks, or independent pharmacies via MediFin/ScriptWise). Revenue is collected from the patient's medical aid scheme at R380 per consultation via Healthbridge.

Patients without medical aid pay via Ozow or PayFast — tiered self-pay pricing that expands Aafiya's addressable market from 9M medical aid members to all 60M South Africans. As NHI becomes operational, capitation payments replace scheme billing as the primary revenue mechanism.

Clinical Safety

Four-Level Escalation Hierarchy

Aafiya's escalation model ensures no patient is stuck at an inappropriate level of care. The system moves cases upward automatically when clinical indicators demand it.

Key principle: The AI can never take the patient down an escalation level. It can only surface cases for physician attention or emergency escalation — never suppress them.

Level 0 — AI Triage: EMERGENCY

Detected by the pre-consultation triage engine before intake begins. Tap-to-call 112/10177 surfaced immediately. No queue — no wait.

Level 1 — ROUTINE / URGENT Queue

URGENT cases surface first in the physician queue with on-call alerts. ROUTINE cases (common cold, minor infections, repeat chronic scripts) follow in order. No live interaction needed for most.

Level 2 — Physician Review

All cases. Physician reviews AI output, approves, modifies, or escalates. Standard workflow. Every action is auditable.

Level 3 — Live Video Consultation

Complex or ambiguous cases escalated by the reviewing physician to a synchronous WebRTC video or audio consultation. Initiated directly from the case review screen.

Level 4 — In-Person / Emergency Referral

Cases requiring physical examination or immediate in-person care are referred to the nearest appropriate facility with a structured clinical handover note.

Access Channels

Designed for Every South African

Aafiya meets patients where they are — not where we wish they were. Multiple access channels ensure no one is excluded by device, data plan, or language.

📱

Progressive Web App

Works on any Android or iOS device. No app store download. Installs from browser. Offline-capable via Workbox service workers.

🎤

Audio Mode

Voice-guided clinical intake for patients with low literacy. Fully functional on 3G connections. No reading required.

🇦🇪

11 Languages

All 11 official SA constitutional languages: English, Zulu, Xhosa, Afrikaans, Sotho, Tswana, Venda, Tsonga, Swati, Ndebele, and Pedi. Including speech-to-text.

💬

WhatsApp Business API

Consultation results, signed notes, and e-prescriptions delivered via WhatsApp in 4 languages. Automatic SMS fallback for patients on low-bandwidth or feature phones.

Chronic Disease Management

Built for South Africa's Biggest Disease Burden

SA has 7.8M people living with HIV, ~4.2M with diabetes, and millions more with hypertension and TB. Aafiya's CDM module turns episodic consultations into continuous longitudinal care.

📋

Care Plans

Structured chronic care plans for hypertension, diabetes, HIV, and TB. Aligned with SA STGs and NIMART protocols. Physician-reviewed and updated at each consultation. Visible to both patient and physician.

📅

Automated Recalls

Recall scheduling built into every CDM care plan. Patients receive WhatsApp or SMS reminders before their next check-in. Overdue recalls surface automatically in the physician queue.

💊

Medication Adherence

Adherence streak tracking for chronic medications. Alerts on missed refills. Integrated with e-prescription routing — patients collect from Dis-Chem or Clicks with a digital code.

⏱

CGM Device Integration

Dexcom CLARITY and FreeStyle Libre continuous glucose monitoring feeds directly into CDM care plans. Apple HealthKit and Android Health Connect adapters planned.

📋

Lab Results — FHIR R4 Framework

HL7 FHIR R4 lab result integration framework with abnormal flag detection. Physicians see results alongside the AI assessment — no manual re-entry, no delays. Partner adapters (PathCare, Ampath) in progress.

🏠

Patient Home Widget

At-a-glance CDM status: days until next recall, latest biometric measurements, medication adherence streak, and upcoming lab due dates — all from the PWA home screen.

🔍

TB CXR AI Pathway

Dedicated TB screening pathway for South Africa's world-high TB burden. AI chest X-ray screening is triggered on TB suspicion during intake or CDM check-in. GeneXpert molecular test results integrate directly into the patient record. MDR-TB flags trigger immediate specialist escalation. Aligned with NIMART protocols and SA TB/MDR-TB guidelines.

CDM as a revenue driver: Chronic patients have higher consultation frequency (4–8 consultations/year vs. 1–2 for acute). At R380/consultation, a cohort of 10,000 CDM patients generates R15–30M/year in scheme billing — with better outcomes and lower emergency utilisation over time.

South African physician reviewing patient data on a digital screen

1,800 unemployed doctors in SA. AI-assisted review enables 15–20 cases/hour.

The Physician Dashboard

Physicians access a structured, time-efficient review interface:

Pre-populated clinical summary from AI intake
Provisional ICD-10 diagnosis with confidence indicators
Proposed treatment plan cross-checked against EML & STGs
Patient history, current medications, and allergy flags
One-click approve, modify-and-sign, or escalate
Audio recording option for complex case notes
Real-time HPCSA registration status indicator

Physician income: R110 per case reviewed (approximately R1,650–R2,200/hour at 15–20 cases/hour). Flexible hours, remote work. Directly addresses SA's 1,800 unemployed qualified doctors.

Physician Partner Model

Empowering SA's Unemployed Doctors

South Africa has approximately 1,800 qualified, registered doctors who cannot find employment in the public or private sector. Aafiya offers these physicians a flexible, remote income opportunity — reviewing 15–20 AI-assisted cases per hour from anywhere with internet access.

This is not a replacement for physicians. It is a force-multiplier: one physician can oversee the equivalent clinical output of 3–5 traditional consultations per hour, with AI handling the data gathering and initial assessment.

Flexible hours — morning, evening, or weekend sessions
Fully remote — no physical clinic required
HPCSA-compliant — every prescription is physician-signed
Continuous CPD — platform tracks learning and feedback

How Aafiya Works