Know which roads are impassable — before an ambulance finds out the hard way.
Indlela Flood combines roadside water-level sensors, rainfall data, vehicle GPS and satellite imagery with multi-model AI to predict which township roads will flood, and turns that prediction into a route advisory a human dispatcher can act on.
No live municipal flood-response system is connected yet. This site and its demo mode use simulated data, clearly labelled.
Real weather snapshot — Alexandra, Johannesburg
Captured 23 Aug 2026: 15.7°C, sunny, 5% chance of rain today. Winter in Gauteng is normally dry — the flash floods described in Research & Evidence mostly happen in the summer rainy season (Nov–Mar).
This is a real, dated reading, not a live feed. The demo below is a simulated scenario, not today's actual conditions.
DEMO · SIMULATED
RD-014 · 8TH STREET BRIDGE, MARLBOROACCESSIBLE
6cm water level · live sim
–60 minrainfall: 3.1 mm/hrnow
Why it exists
A closed road isn't a coordinated response.
When rain hits a township, water rises fast on low-lying roads and public transport, ambulances and residents find out by getting stuck. Indlela Flood closes that gap with sensor evidence, forecasting the roads that will need a detour before they're impassable.
Sense
Roadside water-level sensors, rainfall data and vehicle GPS traces feed the telemetry pipeline continuously.
Predict
Azure ML forecasts road accessibility from current and forecast conditions — flood forecasting, route optimisation, geospatial classification.
Decide
GPT, Claude and Gemini assist a human dispatcher; only a person issues a route advisory.
Road at a river crossing, informal settlementILLUSTRATION
Advisory reroutes traffic around floodingILLUSTRATION
Sensor sends data to the cloud and AIILLUSTRATION
These are original illustrations made for this site, not photographs of a real deployment. See for real news reports this design is based on.
Coverage model
Alexandra to the Cederberg
Indlela Flood is built as a controlled flood-response pilot, expanding zone by zone as sensor coverage and municipal integration are confirmed — not a claim of existing deployment.
Research & Evidence
Why we built this, in plain English
This page explains the problem, how the AI and sensors are meant to help, and shows the real news reports we used to understand the problem. No made-up sources — every link below is a real article.
1. The problem
When it rains hard in South Africa, low-lying roads near rivers can flood fast. People, taxis, and ambulances can get stuck or trapped. This has already happened many times — real floods have closed hundreds of roads and, in the worst cases, cost lives. Right now, most areas only find out a road is flooded once someone is already stuck on it or a resident calls it in.
2. The AI + IoT solution — how each part helps
In simple terms: sensors watch the water, AI checks if it's getting dangerous, and a person decides what to tell drivers.
SENSORS + RAINFALL + GPS + SATELLITE
Small devices on the roadside measure how high the water is. Rain gauges and satellite pictures add more clues. Cars moving slower than normal is another clue.
AZURE ML
This is the part that looks at all the clues together and guesses: will this road still be safe to drive on soon? It's a forecast, not a certainty.
GPT
Turns the forecast into a short, plain message: which road, why, and what detour to use.
CLAUDE
Checks old records: has this exact road flooded before in weather like this? That history makes the warning more trustworthy.
GEMINI
Looks at photos — from satellites or roadside cameras — to see if water is actually visible on the road.
GPT IMAGE 2
Draws a simple picture of the detour, so it's easy to understand at a glance.
HUMAN DISPATCHER
A real person always makes the final call before any warning goes out. The AI never decides alone.
3. See a demonstration
The easiest way to understand this is to watch it happen. Two ways to try it:
A storm around Mthatha, Eastern Cape produced floodwaters several metres deep, swept away vehicles and homes, and led to a national disaster declaration.
Homes right next to the Jukskei River in Alexandra, Johannesburg regularly face rising water in the rainy season, with residents describing having to move to higher ground.
A provincial road (MR310) was washed away by river flooding in 2023 and again in 2024, cutting off a town until it was rebuilt.
3D Model
Play with it — drag to rotate, use the slider
This is a simple 3D model of one road. It is not a real place. Drag your mouse on the model to look around it. Move the slider to add more "rain" and watch what happens.
DryHeavy storm
What you're looking at
ACCESSIBLE
The road — the flat dark strip in the middle.
The sensor — the small pole. It measures how deep the water is getting.
The cloud — shown as the floating shape. The sensor sends its readings here.
The water — turn up the slider and watch it rise over the road.
The vehicle — it will stop and take the side detour once the road is too deep to cross.
This model is simplified on purpose, so it's easy to follow. The real system works the same way — sensor, cloud, AI check, human decision, detour — just with real roads and real sensors.
The Problem
From rising water to a rerouted ambulance
Without sensor evidence, dispatchers usually learn a road is flooded when a vehicle is already stuck on it. Indlela doesn't remove the need for driver judgement — it shortens the distance to a warning.
HEAVY RAINFALL
→
ROAD WATER LEVEL RISES
→
ROAD BECOMES IMPASSABLE
→
EMERGENCY / TRANSPORT DELAYED
→
DISPATCH INVESTIGATION
→
ROUTE ADVISORY
→
SAFE REROUTE
Classification
Local, area-wide, or unknown
Possible Local Closure
One road segment shows rising water while nearby monitored roads stay accessible.
Possible Area Flooding
Several road segments in the same catchment rise together.
Insufficient Evidence
Sensor coverage or data quality is too limited to classify confidently.
An AI accessibility prediction is a lead for a route advisory, never a certified road-safety determination.
How Indlela Flood Works
Sensor to advisory, with a dispatcher at the decision point
WATER-LEVEL SENSOR + RAINFALL + GPS + SATELLITE
→
TELEMETRY PIPELINE
→
AZURE ML ACCESSIBILITY PREDICTION
→
GPT / CLAUDE / GEMINI
→
HUMAN DISPATCH REVIEW
→
ROUTE ADVISORY + GPT IMAGE ROUTE MAP
→
REVERIFICATION AS WATER RECEDES
See it in motion
Run the demo scenario
The Flood Response Operations Centre includes a full simulated walk-through — from rising water to a closed, reverified advisory — using clearly labelled demo telemetry.
AI Intelligence
Each model does one job — a dispatcher decides
Azure ML
Flood forecasting, route optimisation and geospatial classification from sensor, rainfall and satellite inputs.
prediction
Latest GPT model
Turns a predicted closure into a route advisory: affected segment, evidence, recommended detour, status.
advisory generation
Latest Claude models
Analyses historical flooding and road-closure records to check whether this pattern has occurred before.
long-context history
Google Cloud Gemini
Analyses satellite and roadside imagery for visible water extent and road condition.
visual evidence
GPT Image 2
Generates a simple emergency route map for the advisory — labelled as AI-generated, never documentary proof of conditions.
AI-generated, labelled
Human review
Every route advisory requires Approve, Modify, Reject, Request Field Check, or Escalate from a dispatcher.
required
Model identifiers are verified against official provider documentation at implementation time and are never hard-coded from memory. This prototype uses simulated AI outputs to demonstrate the workflow shape, not live model calls. Estimated cost model: R5,000–R20,000/month per municipal flood-response system — a project estimate, not a guaranteed price.
Deployment Zones
12 real flood-risk locations we're planning around
These are real places in South Africa with a real history of flooding or road closures (see Research & Evidence for sources). The weather numbers below are real and live — pulled from a public weather service each time you load this page. The zone status is our internal planning stage, not a confirmed deployment. No road-level water sensors are installed at any of these zones yet — that part stays simulated until real hardware is on site.
Live weather via Open-Meteo · auto-refreshes every 5 minutes
About Us
Phakama Nkosazana NPC
Indlela Flood is developed under Phakama Nkosazana, a registered South African non-profit company building AI-driven public-safety and social-impact tools. Our registration is public record with the Companies and Intellectual Property Commission (CIPC).
Registration
Enterprise Name: Phakama Nkosazana Registration Number: 2023 / 277292 / 08 Enterprise Type: Non Profit Company Registration Date: 21/12/2023 Status: In Business
Registered Office
34 Squirrel Street Villa Liza Boksburg Gauteng, 1459 South Africa
Tax Reference
Tax Number: 9203143277 Financial Year End: February
Contact
Departmental email addresses
General
admin@phakamankosazana.co.za
Finance
finance@phakamankosazana.co.za
Developers
developers@phakamankosazana.co.za
AI Solutions
aisolutions@phakamankosazana.co.za
Requests for Quotation
rfq@phakamankosazana.co.za
Leadership
Our Directors
Agisanang Samkeliso Rakgotho
Director
Email: agisanang@phakamankosazana.co.za
Cell: 073 567 7670
"My inspiration comes from watching communities cut off — from help, from safety, from opportunity — simply because no one saw the risk coming in time. My goal with Phakama is to change lives by putting early warning in the hands of the people who need it most, not just the institutions that can afford it. What Phakama will do for our people is simple: turn data that already exists into decisions that save time, and sometimes save lives. I believe in this vision because I've seen what happens when technology is built with a community instead of at a distance from it — it stops being a tool and starts being trusted."
Chantelle Palesa Malahlela
Director
Email: chantelle@phakamankosazana.co.za
Cell: 081 306 7945
"I started Phakama because I grew tired of watching preventable disasters get treated as unavoidable. My goal is to change lives by making sure the people most exposed to risk are the first to be protected, not the last to be informed. What Phakama will do for our people is give communities a voice inside systems that were never built to listen to them — turning silence and delay into warning and action. I believe in this vision because uplift, 'phakama', has to mean something real: safer roads, faster responses, and technology that serves dignity before it serves convenience."
Sign in to the Operations Centre
Demo access only. No account is created or verified — this opens the simulated dispatcher experience.
DEMO MODE — ALL TELEMETRY, PREDICTIONS AND AI OUTPUTS ON THIS SCREEN ARE SIMULATED
Flood Response Operations Centre
Signed in as Municipal Flood Response Manager · simulated session
10
Monitored road segments
0
Impassable segments
0
Active advisories
0
Vehicles rerouted (est.)
System status
Last telemetry update: just now
AI service status: Operational (simulated)
Sensor health: 9 of 10 online
Recent alerts
--:--No alerts yet. Run the demo scenario to generate one.
Road Status Map
Alexandra & Marlboro, Johannesburg · real place names, simulated readings
These are real Johannesburg locations known for flooding near the Jukskei River, used here to make the demo realistic. No sensors are actually installed at these roads yet — every number below is simulated. Colours are simple to read: ACCESSIBLE safe to drive, CAUTION water rising, watch it, IMPASSABLE do not cross, SENSOR OFFLINE no reading available.
RD-014 · 8th Street Bridge
Marlboro · monitoring active
Water level — last 60 min (simulated)
Current: 6 cmRainfall: 3.1 mm/hrImpassable threshold: 25 cm
Sensors
Water-level sensor: OK · signal strong
Rainfall gauge: OK · signal strong
Connectivity: LTE-M · last seen 4s ago
Historical closures
No prior closures logged for this segment yet.
Advisory & audit history
No prior advisories recorded on this segment.
AI Route Advisory
Requires human dispatch review before any advisory is issued
No AI route advisory has been generated yet. Run the demo scenario to trigger one from a predicted closure.
Advisories Issued
Created only after human approval of an AI recommendation
No advisories yet.
Dispatcher View
Mobile-first advisory flow (desktop preview)
My tasks
No assignments yet.
Route map
No route map generated. Assign one from the demo scenario.