Language changes apply instantly; translations are drafts for review by a fluent speaker. Unlisted text shows in English.
Water level rising: sensors report, a person decides

KwaZulu-Natal and Eastern Cape · community flood early warning

Tell the village before the river rises, in the language they speak.

River level, rainfall and photo readings from solar sensors go in. A scored risk, a drafted alert in the community's language and a work order for a person to approve come out. Municipal and traditional leaders see the same picture.

River level
1.42 m Pilot test data
Rain, next 24 h
18 mm Pilot test data
Sensors online
5 of 6 Pilot test data

This solution requires current AI models to operate.

Why floods become disasters

Heavy rain falls upstream
No gauge watches the river
Level rises unseen
Warning arrives late or not at all
Alert is in the wrong language
Leaders cannot confirm quickly
People move late
Homes, crops and lives are lost

These eight steps follow the 2022 KwaZulu-Natal and 2025 Eastern Cape flood records on the Evidence page.

Silent rivers

Flood-prone settlements sit mostly in low-lying areas beside rivers and below steep slopes (Evidence E6), and few of those rivers have a gauge anyone is watching.

Late alerts

Warnings are issued nationally but often do not reach, resonate with, or move the people at risk to act (WMO, Evidence E10).

Left-out leaders

Municipal officers and traditional leaders need the same picture at the same time. The 2025 national disaster committee included both.

Decision support only. Responsibility stays with the humans and authorities who approve action.

How it works: the Scenario Theatre

Speed

Human decision needed

Run results Pilot test data

ScenarioTriggerCaught byHuman decisionMinutes to decision

AI helpers and cloud helpers

Every AI and cloud service below runs on Amazon Web Services (AWS). A named person approves every alert.

How they talk to each other (3D)

This solution requires current AI models to operate. The Amazon Bedrock foundation models listed (or their current equivalents) are essential to the production system.

3D model

3D needs WebGL, which this browser does not offer.

20% STABLE

Click a part to learn what it does.

How this model works

The model shows a village beside a river. A solar pole measures how high the water is and sends the number to a phone. Slide the river up and the badge changes from STABLE to ATTENTION to HIGH RISK, and the phone screen changes colour with it. Press "Show parts apart" to see each piece with a plain label, then click a piece to learn what it does. In real life a person, not the machine, decides whether to send the warning.

Impact

Why TUS INTELLIGENCE is building Umlindi

Floods in KwaZulu-Natal and the Eastern Cape show a gap between a warning being issued and a community acting on it. TUS INTELLIGENCE is building Umlindi to close that gap with river-level readings, plain-language drafts in the community's own language, and a named person who approves every alert.

What the records show

The April 2022 KwaZulu-Natal floods killed 435 people, left 80 missing and affected 128,743 people, with 600 schools and 84 health facilities damaged (IFRC, E1). In the Eastern Cape floods of June 2025, 101 deaths were confirmed by 26 June (IOL, quoting the provincial MEC for CoGTA) and a national disaster was declared. These figures describe the problem, not our results.

What the pilot will measure: minutes from first anomaly to approved alert, share of alerts a human confirms, work orders verified closed. No pilot results exist yet and none are claimed.

Residents

A warning in their own language, with a plain action, before the water reaches the home.

Traditional authorities

The same river readings and drafted alert the municipality sees, so they can confirm and relay it.

Municipal disaster officers

A ranked queue, one approval screen, and an audit trail of who decided what and when.

Schools and clinics

Early notice when the road or river beside them is rising, so learners and patients can be moved first.

Why we think it can work

Warnings already flow from the South African Weather Service through provincial and local channels, and a person already decides what to tell a community. In Quarry Road, Durban, residents' leaders tied red tape along the riverbank on 11 April 2022 as the river rose (E5). Umlindi gives that same human judgement a river reading, a draft and a record.

R&D panel

Hypotheses: (1) river-level sensing plus a human gate shortens the time from first anomaly to approved alert; (2) cross-checking neighbouring sensors keeps false alarms low. Measures: time to decision, share of alerts confirmed, work orders verified. Method: log every event in the workspace audit trail. Publication: results are published only after the pilot has collected them.

Evidence

In April 2022, record rain fell around Durban and 435 people died, most in informal settlements beside rivers or on steep hillsides. In June 2025, floods around Mthatha in the Eastern Cape killed more than 100. Both disasters happened in a country with a national weather service that issues warnings. The records show a gap between a warning being issued and a community acting on it: messages that did not reach people, did not make sense to them, or did not come from someone they trusted. In Quarry Road, Durban, a community-based system built with UKZN let residents' leaders mark the riverbank as the water rose. The Auditor-General and the Human Rights Commission also found that the state's response afterwards was too slow. Umlindi targets the gap: a solar sensor watches the river, AI drafts a plain-language alert in the community's language, and a named person approves it.

Why this solution is needed
FindingCardsUmlindi capability
Informal settlements beside rivers bear the harmE1, E2, E6Solar river-level sensors at those sites
Warnings exist but do not always reach or move peopleE5, E7, E9, E10Local-language drafts approved by a person
Community and traditional leaders matterE5, E6Authority and community views
Networks fail in the stormE2Edge buffering and SMS fallback
State response afterwards was slowE3, E4, E8Decision queue, work orders, audit trail

Together the ten sources show the scale, the mechanism and the gap. They do not show that Umlindi works; the pilot will confirm that with its own measurements.

10 problems people have told us about

Coverage

Live rainfall from Open-Meteo. Loading…

AreaTypeStatusRain now (mm)Rain next 24 h (mm)

Status shows planning priority only. Priority pilot areas are where the company intends to begin; no installed-sensor results are claimed.

About

Umlindi by TUS INTELLIGENCE.

Registration

Registered name: TUS INTELLIGENCE
Registration number: 2026/911684/07 (private company)
SARS income tax reference: 9001341370, registered 12 November 2026
Registered address: 945 Pinehurst Place, Copperleaf Golf Estate, Centurion, Gauteng, 0157, South Africa

admin@tusintelligence.co.za
finance@tusintelligence.co.za
engineering@tusintelligence.co.za
rfq@tusintelligence.co.za
contact@tusintelligence.co.za
Thandolwethu Unathi Sibaya, Director

Director of TUS INTELLIGENCE, responsible for the Umlindi flood early-warning product. Umlindi is decision support: a named person always approves any alert before it is sent.

Contact

Sent from your own email app. Nothing is stored on this site.

Thank you. Your email app should now be open. Nothing is stored on this site.
PILOT STAGE: LIVE TESTING PRODUCT View as

Pilot testing access