Know which roof is about to leak — before it damages the ceiling below.
RoofGuard SA reads roof moisture sensors, rainfall data and inspection imagery, turns them into an AI-drafted maintenance recommendation, and puts it in front of a facilities manager for a human decision — every time.
DEMO — ROOF MOISTURE (CITY HALL, JHB)
38%
moisture reading vs 65% leak-risk threshold
Status: NORMAL · demo data, refreshes every few seconds
Illustration — moisture sensors across the roof deckIllustration — AI recommendation routed to a human
Live today as a working prototype across 6 reference buildings; the roadmap runs from these 6 buildings to every municipal property portfolio with a deferred-maintenance and roof-leak problem.
The Problem
Roof leaks stay invisible until the ceiling comes down.
South Africa's public infrastructure maintenance backlog runs into the tens of billions of rand, and roof and water-ingress damage is a recurring, largely unmonitored contributor — by the time a leak is visible from inside a building, the ceiling, wiring and internal facilities are already compromised.
NORMAL WEATHERING
→
MOISTURE SENSOR SIGNAL
→
THRESHOLD BREACH
→
CEILING / ELECTRICAL DAMAGE RISK
→
FACILITIES MANAGER INVESTIGATES
→
INSPECT / PATCH / ESCALATE
→
ROOF RE-VERIFIED
Isolated Wear Point
One flashing or membrane seam is drifting out of spec while the rest of the roof is fine — a scheduled patch, not an emergency.
Building-Wide Pattern
Multiple zones of the same roof show the same drift after a storm — points to a structural or membrane-batch issue needing a full inspection.
Insufficient Evidence
Sensor readings are sparse or conflicting — the system flags "needs physical inspection" rather than guessing.
Every RoofGuard output is a maintenance lead for a human decision-maker — never an automatically issued instruction and never a certified structural determination.
How It Works
From a moisture drift to a repair booking, every step is logged.
The pipeline behind RoofGuard SA — the same one the interactive demo scenario runs end to end.
MOISTURE SENSORS + RAINFALL + INSPECTION IMAGERY
→
DATA INGESTION
→
ML ANOMALY FORECAST
→
LLM ANALYSIS (MULTI-MODEL)
→
HUMAN REVIEW
→
MAINTENANCE RECOMMENDATION
→
RE-VERIFICATION LOOP
AI Intelligence
Six distinct components, each doing one job.
No single model runs the whole pipeline — each stage is a separate, auditable step, with a human always at the end.
Forecasting
Leak Pattern Model
Azure ML — turns moisture-sensor and rainfall data into a leak-pattern anomaly score per roof zone.
Advisory LLM
Maintenance Recommendation
Latest GPT model — converts the anomaly score into a short, human-readable maintenance recommendation with reasoning.
Historical Pattern
Prior Repairs Review
Latest Claude model — checks the current reading against records of previous roof repairs on this building.
Vision
Roof Photograph Analysis
Google Gemini — analyses inspection photographs to visually confirm membrane cracking, ponding, or flashing damage.
The mandatory final decision-maker on every recommendation. No work order is issued without this step.
Model identifiers above should be verified against each provider's current documentation rather than assumed from training data. This build's AI outputs are simulated to demonstrate the workflow shape, not live model calls. Cost figures elsewhere on this site are labelled estimate ranges only.
The Interactive Model
Drag to look around. Slide to soak the roof.
A simplified, to-scale simulation of one building roof: deck, a moisture sensor, the AI pipeline, and the facilities manager's decision point.
DRY
Roof moisture within normal operating range.
What you're looking at
01 The sloped deck is the building's roof — where moisture sensors are installed under the membrane.
02 The pulsing pole is a moisture sensor, reporting saturation in real time.
03 The floating shape is the AI pipeline — forecasting, cross-checking and drafting a recommendation.
04 The panel is the facilities manager's decision point — nothing is actioned without it.
05 The small block is a maintenance cart. Past the critical threshold it moves to the affected zone instead of continuing its routine round.
This model is simplified for clarity. The real system follows the same shape — sensor → AI → human — using real hardware and real rainfall data.
Research & Evidence
Why this matters, in plain English.
South Africa's public infrastructure maintenance backlog is measured in the tens of billions of rand, and water damage from unmonitored leaks is a well-documented part of that story — this isn't a hypothetical problem.
Moisture sensors + rainfall + inspection imagery
Tell the system what's actually happening on the roof right now, instead of relying on the next scheduled inspection to catch a problem.
The leak-pattern forecasting model
Turns those raw readings into a score: how likely is this roof zone to develop a leak in the near future.
The advisory and historical-pattern LLMs
Turn that score into a plain-English maintenance recommendation, checked against this building's repair history.
The vision model
Double-checks the recommendation against an actual inspection photo, so the system isn't guessing blind.
The human decision-maker
Reads all of the above and decides — approve, change, reject, or send someone to inspect in person.
Reports a public works minister acknowledging a R30 billion maintenance backlog affecting over 56,000 state-owned properties, including thousands of government buildings needing refurbishment.
States that the municipal infrastructure backlog across South Africa now stands at roughly R1 trillion, driven by capital spending falling steadily as a share of municipal budgets.
Notes that rainstorms repeatedly trigger further infrastructure failures in Johannesburg, on top of a maintenance and upgrade backlog running into the hundreds of billions of rand.
A property-law specialist reports that roof and water leaks make up roughly half of her department's case load, and describes them as one of the biggest threats to a building's asset value.
Covers a business chamber's call for Nelson Mandela Bay municipality to prioritise its escalating infrastructure maintenance backlog, including water leak repairs, in its annual budget review.
A construction industry report documenting how South African municipalities consistently under-budget for infrastructure maintenance, allowing deterioration to compound over time.
Deployment / Coverage Zones
6 reference buildings, live rainfall feed.
Rainfall is the single biggest driver of roof leak risk — the cards below pull genuinely live, auto-refreshing rainfall and precipitation data for each building from Open-Meteo (no API key, refreshes every 5 minutes).
About & Team
ZYRE
Legal entity details below reflect the CIPC registration certificate and SARS notice of registration.
Municipal buildings across South Africa are losing the fight against water damage one unnoticed leak at a time — by the time a ceiling stain appears, the real cost is already locked in. RoofGuard SA exists to give facilities teams the one thing they never have: advance warning. Our goal is simple but urgent — catch the leak before it happens, not after, so municipalities spend their maintenance budgets on prevention instead of emergency repair. Done right, this doesn't just save money; it protects the schools, clinics and public buildings that communities depend on every day, and keeps them open, dry and safe for the people who use them.
Gated Demo
Enter the Operations Centre
Select the role you'd be evaluating this as.
Demo access only. No account is created or verified. Every value inside the Operations Centre is simulated.
DEMO MODE — ALL TELEMETRY, PREDICTIONS AND AI OUTPUTS ON THIS SCREEN ARE SIMULATED
Operations Centre
6
Monitored Buildings
1
Zones At Risk
1
Active Recommendations
0
Work Orders Issued (Session)
System Status
Last sensor updatejust now
AI serviceOperational (simulated)
Sensor / device health96 / 104 online
Roof zones under active alert1
Recent Alerts
08:41City Hall (JHB) north-east roof zone crossed 65% moisture
08:22Pretoria Civic Centre moisture trending up ahead of rain
07:55Durban Municipal Offices sensor cluster reconnected
Status / Asset Grid
Roof Zone Detail — City Hall, Johannesburg (NE Quadrant)
Source ModelGPT (advisory) + Claude (repair history)
Affected ZoneCity Hall, Johannesburg — NE Quadrant
Evidence71% moisture, rising trend, 22mm rain in 24h
RecommendationDispatch crew — inspect and patch NE membrane
ConfidenceHigh (0.88)
Issued Actions / Work Orders
No actions issued yet this session. Approve a recommendation in "AI Recommendation" or run the demo scenario to populate this list.
Field / Maintenance View
Today's Tasks
No tasks assigned yet. Tasks appear here once a recommendation is approved.
Run Demo Scenario
Fires the full pipeline for City Hall, Johannesburg (NE Quadrant). Pauses at the AI recommendation stage for a human decision — it never auto-approves.