IsakhiwoWatch
6 monitored public housing clusters · South Africa

Know which crack needs an engineer — before a wall does.

Isakhiwo Watch turns periodic smartphone photographs and crack-width sensor readings into a plain-language structural risk assessment that an engineer reviews and approves before a household is told anything.

Cluster Snapshot Demo preview
0.6mm
Crack Width Growth Rate — Savannah City Block 4, De Deur
Status: Widening — assessment drafted, pending engineer review

Not live data — see the Coverage Zones page for the genuinely live weather-conditions feed.
ENG
ILLUSTRATION — a photographed crack travels through the AI pipeline to an engineer's dashboard
0.6mm
ILLUSTRATION — the crack-growth gauge an engineer sees before approving an inspection request
Starting with subsidised housing developments flagged for structural complaints in Gauteng, expanding toward every public housing portfolio in South Africa.
The Problem

Cracks get dismissed as cosmetic until they aren't.

A hairline crack behind a door frame can be nothing, or it can be the first visible sign of foundation movement. Without regular measurement, residents and officials have no way to tell the difference until the crack is already wide enough to see from across the room.

CRACK
FORMS
WIDTH
DRIFTS
THRESHOLD
BREACH
RESIDENT
CONCERN
ENGINEER
INVESTIGATION
REPAIR
DECISION
STRUCTURE
STABILISED

How ambiguous situations get classified

Case A

Cosmetic / Stable Crack

Width is within the 0.08mm hairline threshold and shows no growth across readings. Logged, not escalated.

Case B

Progressive Structural Crack

Width is growing consistently across readings — a sign of ongoing movement. Escalated for an engineering visit.

Case C

Insufficient Data

Photos are inconsistent in angle or lighting, or sensor readings conflict. The system flags "needs re-photograph" rather than guessing.

In every case, the AI output is a lead for a qualified engineer to investigate — never a certified determination of structural safety.

How It Works

From a smartphone photo to an approved inspection order.

Every stage below is real in architecture. In this prototype, the imagery and AI outputs are simulated so you can experience the workflow before any hardware is deployed.

SMARTPHONE PHOTOS
+ CRACK SENSORS
IMAGE
INGESTION
AZURE ML
CHANGE TRACKING
MULTI-LLM
ADVISORY LAYER
HUMAN
ENGINEER REVIEW
INSPECTION
ISSUED
RE-VERIFICATION
LOOP
AI Intelligence

One job per model. No single AI does everything.

Vision

Google Gemini

Analyses crack imagery from smartphone photographs to measure apparent width, length and orientation.

Change detection

Azure ML

Tracks how each crack changes across repeated photographs and sensor readings over time.

Summary drafting

Latest GPT model

Converts the change-detection output into a short, human-readable inspection summary.

Historical pattern

Latest Claude model

Reviews historical structural reports for the same housing cluster for precedent.

Generative asset

GPT Image

Produces an illustrative crack-monitoring diagram showing where measurements were taken, explicitly labelled AI-generated.

Required

Human Structural Engineer

The mandatory final decision-maker. No household is told a wall is unsafe without their sign-off.

Model identifiers should be verified against each provider's current documentation rather than assumed. This build's AI outputs are simulated to demonstrate the workflow shape, not live model calls. Estimated cost: R1,500–R7,000 per month, per building cluster — a labelled estimate range, not a quote.

Interactive Model

Drag to rotate. Move the slider to see a wall react to crack growth.

WIDENING — MONITOR

The crack is measurably wider than the previous reading. Not yet urgent, but worth a follow-up photograph next week.

  1. Clay pole: a crack-width sensor reporting its measurement every few seconds.
  2. Blue icosahedron: the AI pipeline — vision model plus the advisory LLMs.
  3. Dark panel: the engineer's decision point — nothing is issued without it.
  4. Crack line on the wall: visibly widens and darkens as the growth rate crosses the critical threshold.

This model is simplified for clarity. The real system follows the same shape, with real smartphone photographs and sensors in place of this demonstration.

Research & Evidence

Why this matters, in plain English.

South Africa has built more than three million subsidised public housing units since 1994, and a large share of them have documented structural defects — some cosmetic, some serious enough to require demolition and rebuilding. Right now, most of that assessment relies on someone noticing a crack and complaining loudly enough.

How each part of the stack helps

  • Smartphone photos & sensors
    Give the system a consistent, repeatable measurement of every tracked crack.
  • Gemini vision analysis
    Reads each photo and measures the crack's width, length and orientation.
  • Azure ML change tracking
    Compares today's measurement to every past one, to see if it's actually getting worse.
  • GPT summary model
    Writes that comparison into a sentence an engineer can act on.
  • Claude history check
    Asks: has this housing cluster had structural problems before, and what caused them?
  • Human decision-maker
    Reads all of the above and decides — approve, adjust, or reject the inspection request.

Real sources

Deployment / Coverage Zones

Six South African metros, tracked today.

This section pulls genuinely live weather conditions for each metro from Open-Meteo — a real, free public API — because temperature swings and rainfall are measurable drivers of soil movement and thermal expansion, both of which affect crack growth. Building-level photographs and sensor readings shown elsewhere on this site remain simulated.

Weather readings above are fetched live from Open-Meteo and refresh automatically every 5 minutes. Each card's "deployment stage" label is Isakhiwo Watch's own internal designation — it is not derived from the live weather feed.
About & Team

A real, contactable organisation.

About Us

Isakhiwo Watch is developed under Anele Dlamini Holdings, a registered South African company building AI-driven structural-safety tools for public housing. Our registration is public record with the Companies and Intellectual Property Commission (CIPC).

Departments

Director

AD

Anele Dlamini

Director · anele@aneledlaminiholdings.co.za · 081 716 1414

"I grew up around houses where a crack in the wall was just something you learned to live with — nobody had the money or the standing to ask whether it meant something worse. My goal with Isakhiwo Watch is to change that: to put the same kind of monitoring a wealthy property gets in the hands of families in subsidised housing, for free to them and paid for by the people responsible for the building. What we'll do concretely is turn a photo taken on an ordinary phone into evidence an engineer can actually act on, so a family isn't stuck choosing between being ignored and being alarmed. I believe in this because dignity starts with being taken seriously, and a home is the first place that should happen."

Demo Mode — All imagery, predictions and AI outputs on this screen are simulated