PILOT STAGE — LIVE TESTING PRODUCT
Flood early warning · Rural villages & informal settlements · South Africa

Warn the river’s neighbours first — in their own language, through people they trust.

RiverVoice by Azhi Holdings turns solar river-level, rain, temperature and camera readings into plain-language WhatsApp and SMS warnings, drafted for review by a person, and shared with municipal and traditional-authority dashboards.

Open demo workspaceSee the 3D model

Inputs

River level · rain · temperature · photo · GPS

Edge AI

Anomaly detection near the river, even when the network is weak

Outputs

Multi-language alerts · authority dashboards · field tasks
The problem

Warnings exist. Reaching the last household is the gap.

Floods in 2026 again hit informal settlements and villages near rivers. Analysts of the January Limpopo floods point to weaknesses beyond forecasting, including how warnings reach people and trigger protective action. Where a warning does not arrive, households carry the health, livelihood and dignity costs.

Heavy rain falls upstream
River level rises unseen
Forecast is regional, not local
No local sensor confirms it
Warning arrives late or in the wrong language
Households and leaders cannot coordinate
Homes, health and livelihoods are hit
Recovery is slow; data is lost, so the cycle repeats

Visibility gap

Nobody sees the local river in real time.

Language gap

Alerts must work in the languages people speak at home.

Trust gap

Traditional authorities and ward leaders need a seat in the loop.

Liability note: RiverVoice supports decisions. Municipal and traditional authorities, and the people on the ground, remain responsible for declaring evacuations.

How it works

From riverbank to household in six steps

  1. Solar sensors read level, rain, temperature and photos, with GPS position.
  2. Edge models flag unusual rise rates on the device.
  3. The cloud AI compares readings with forecasts and history.
  4. Generative AI drafts alerts in the chosen languages.
  5. A human reviewer approves, modifies, rejects or escalates.
  6. Approved alerts go out by WhatsApp/SMS and appear on authority dashboards.
Run a scenario
AI helpers

Each helper has one job. A person has the last word.

Edge anomaly detector

Small model on NVIDIA Jetson-class hardware flags abnormal rise rates. Cost range: low, device-only.

Reasoning model (GPT-class)

Combines sensor, forecast and history signals into a structured recommendation. Cost: per-request, low at pilot volume.

Alert writer

Drafts WhatsApp/SMS text in several South African languages in simple words. Needs local-speaker review.

Image checker

A vision/image model reviews sensor-camera photos for visual confirmation of water on roads, crossings or yards.

Coordination drafter

Prepares briefing notes for municipality, ward committee and traditional authority.

Human reviewer (mandatory)

Approves every alert. No alert is sent automatically.

Model names change quickly. The pilot uses whichever current GPT-class text and image models are available on the chosen cloud (Azure OpenAI or AWS Bedrock) at deployment time.

SensorsEdge AIReasoning + ImageAlert writerHuman decides
3D model

See how the pieces fit together

Drag to turn. In assembly view, tap a part.
Impact

Impact and founder’s message

“Floods do not wait for good phone signal or for the right language. When I look at the villages and settlements along our rivers, I see families who were never told in time, or were told in words they could not use. Azhi Holdings exists to fix that. We read the river, we write a clear warning in the language people speak at home, and a trusted person decides before it is sent. Technology gives us speed. People give it trust. I want every household near a river to hear the warning early and know what to do.”
— Oratile Mohale, Owner and Director, Azhi Holdings

Only sourced facts are shown here; pilot results will be added once measured:

Research & evidence

Sources

Articles

Ten problems, ten ways the system helps

Coverage

20 areas

Status labels show pilot plans. No live readings are displayed until a real API connection is configured.

About

Azhi Holdings

Registration

AZHI HOLDINGS, registered company
Reg. no. 2026/882684/07
989 Thornridge Estate, Centurion, Gauteng, 0157

What we do

RiverVoice flood early warning: solar river sensors, edge AI, and human-approved WhatsApp/SMS alerts in local languages.

Pilot stage

RiverVoice is a live testing product. A person approves every alert.
Owner

Meet the owner

Oratile Mohale

Owner and Director, Azhi Holdings
oratile@azhiholdings.co.za
Email the owner
Departments

Who to write to

Info

General questions about RiverVoice and pilots.
info@azhiholdings.co.za

RFQ

Requests for quotation and tenders.
rfq@azhiholdings.co.za

Admin

Company records, contracts and general administration.
admin@azhiholdings.co.za

Finance

Invoices, payments and statements.
finance@azhiholdings.co.za
Tax

SARS tax registration certificate

Official SARS Notice of Registration for Azhi Holdings.

Registered name: AZHI HOLDINGS
Registration no.: 2026/882684/07
Taxpayer reference no.: 9780061207
Date of notice: 2026-11-03

Contact

Write to us

Demo workspace

RiverVoice pilot workspace

Pilot testing access

PILOT STAGE — LIVE TESTING PRODUCT. Telemetry, predictions and AI outputs here are pilot-test values.