Pilot scope · Household air monitoring · South African settlements
Know when the air at home turns dangerous.
Low-cost sensors measure PM2.5, carbon monoxide and VOCs inside and outside homes. Edge AI spots harmful episodes, drafts simple advice for families and maps risk for health workers. People review every alert and every message.
Dashboard mockup · illustrative, not live
PM2.5 indoor0
CO ppm0
VOC index0
Sensor integritySIGNEDdesign intent
Illustration: an indoor sensor, an outdoor sensor and rising smoke. Original drawing, not a photograph.
Illustration: a health risk map with darker cells for higher exposure. Original drawing.
What this system does not do: it does not diagnose illness, fine or penalise households for burning fuel, or share household readings without consent. It raises alerts and drafts advice. People decide. AI dependency: This solution requires current AI models to operate.
01 · The problem
Smoke you cannot see, air you cannot choose
Where paraffin, coal, wood and waste are burned for warmth and cooking, indoor air can turn harmful without any warning. Families often lack the tools to see it, and clean options may be out of reach.
Cold eveningBurn fuelSmoke indoorsExposureCough & illnessClinic & missed schoolClean-energy support
Normal cooking spike
Short rise that clears. A gentle tip is enough.
Repeated evening episodes
Same home or block, most nights. Flag for a health worker visit.
Hazardous episode
High carbon monoxide or PM2.5. Alert the household and a human right away.
Liability: outputs are alerts and leads for human review, never medical diagnoses or certified measurements.
02 · How it works
From sensor to safer home
03 · Intelligence roles
One job per model. A human at the end.
IoT sensors
PM2.5, CO and VOC sensors indoors and outdoors. Hardware, not AI.
GPT-6 Luna · Episodes
Detects hazardous episodes and separates them from normal spikes.
GPT-6 Sol · Advice
Drafts short, plain behaviour advice for families.
GPT-6 Astra · Risk maps
Builds health risk maps from approved readings.
Image 2 · Burning sources
Identifies likely burning sources (stove, brazier, waste fire) from photos of the site.
Human reviewer · Mandatory
A community health worker or municipal air quality officer reviews every alert and message before it is sent.
Disclosure: model names are as supplied in the project brief and must be checked against current provider documentation. All AI outputs in this prototype are simulated to show workflow shape. Outputs are probabilistic and not medical advice. Monthly cost range: to be confirmed. This solution requires current AI models to operate; the models listed (or their current equivalents) are essential to the production system.
04 · Interactive model
See the air, then act
Clean
Drag to rotate · scroll or pinch to zoom · simplified for teaching · levels are teaching thresholds, not health limits
Clear walls: the home. Dark cylinder + grey puffs: a stove and its smoke. Ball on the back wall: indoor sensor. Ball on the pole: outdoor sensor.
Wire shape: the AI watching. Phone: advice screen; it turns amber, then red. Blue cube: a family member following the advice. Stays put when clean, moves to the open window when smoky, steps outside when hazardous.
05 · Research & evidence
Why this is needed
South African studies link household burning of paraffin, coal and wood with breathing problems. In one Western Cape study of 590 schoolchildren in informal settlements, paraffin cooking, damp and passive smoking were each linked to roughly two to three times the risk of airway problems. These sources were found by search. Open each one and confirm it before launch.
Five sources listed. Open each link and confirm it before launch.
06 · Impact
Why this matters
A note from the founder
“I grew up knowing how cold evenings push families to burn whatever they have to stay warm, and how invisible that smoke is. I started Zephyra Mind AI to give every household a way to see the air they breathe, and to give clinics and municipalities the early warning they need.” — Kitso Tshepo Joseph Mofokeng, Director
Families
See when air turns harmful and get plain advice in time to act.
Community health workers
Visit the homes that need it most instead of guessing.
Municipal air quality officers
Risk maps show where exposure clusters, to plan support.
Clinics
Early warning of bad evenings helps prepare for respiratory cases.
Funders
Evidence to target clean-energy support where it helps most.
Why we think it can work: community health workers already visit homes and decide. We add earlier, clearer signals; a person still approves every alert and message. Honest scope: these are design intentions for a prototype, not measured results.
Director's note: “I grew up knowing how cold evenings push families to burn whatever they have to stay warm, and how invisible that smoke is. I started Zephyra Mind AI to give every household a way to see the air they breathe, and to give clinics and municipalities the early warning they need. My goal is simple: fewer coughing children, fewer missed school days, and a cleaner, safer home for every family in our settlements.” Kitso Tshepo Joseph Mofokeng, Director
Investment: Clean Air Alert is a product of Zephyra Mind AI. The startup currently has no external funding and requires investment to begin the new project and move from this working prototype into production. Microsoft Azure for Startups, AWS Activate / AWS Startups, NVIDIA Inception, and peer AI and cloud startup programmes are the natural partners for this stage because Clean Air Alert is built on the same stack those programmes exist to accelerate: modern AI models for scoring, advisory generation, long-context pattern analysis and vision verification, plus cloud infrastructure for secure data ingest and human-in-the-loop decision support. Grant or programme support from these partners directly funds the next phase — real backend, real integrations, and pilot deployments with municipalities or enterprise buyers — while the human-always-decides architecture already demonstrated on this site keeps governance and liability clear. Without that support the project remains a high-fidelity prototype; with it, Zephyra Mind AI can deliver a production system that depends on current AI models and that those programmes are designed to help scale.
