Solar IoT sensors on boreholes, tanks and standpipes feed edge anomaly detection. GPT-class models draft multi-language WhatsApp and SMS alerts. Image analysis confirms visual status. Coordinated drafts go to municipal teams and traditional authorities. Humans always decide the next action.
Borehole pumps fail, transformers trip, pipes are removed, and tankers arrive late or not at all. Communities in Limpopo villages and Gauteng informal settlements walk kilometres or buy water while children fall ill. Visibility is missing; response is delayed; dignity is eroded.
Ageing pumps, stolen cables, collapsed boreholes documented across Limpopo and Eastern Cape villages.
Municipal dashboards rarely show individual community points; traditional authorities lack shared status.
Tankering bills reach hundreds of millions while permanent fixes remain unfunded or delayed.
This product generates leads and decision-support drafts only. Liability remains with the operating water services authority and human decision-makers.
Models under evaluation for this product. All outputs require human review before any external communication or work order.
Lightweight model on solar node or gateway. Detects flow/level/power anomalies. Cost range: low (edge inference).
GPT-class / Claude / Gemini candidates generate WhatsApp and SMS text in Sepedi, isiZulu, Afrikaans, English. Marked candidates under evaluation.
GPT Image 2.5 and successors or equivalent vision models analyse photos of water points. Candidates under evaluation.
Generates structured notes for municipal engineers and traditional leaders. AWS Bedrock / Azure OpenAI / Vertex candidates.
Mandatory. Reviews every recommendation, alert draft and work-order suggestion. No auto-approve path exists.
Latest relevant models under evaluation include GPT-4o / GPT-5 class, GPT Image 2.5, Claude Opus/Sonnet/Haiku, Gemini 1.5/2.x, NVIDIA NIM / TensorRT for edge, AWS Bedrock, Azure OpenAI. All marked candidates under evaluation.
How the models work together: 1. Sensor data arrives at the edge node. 2. Anomaly model flags an issue and packages a summary. 3. Alert drafter and image model receive the package in parallel. 4. Both return drafts to the coordination layer. 5. Human reviewer sees the combined recommendation and decides. 6. Only after the human click does any external message or work order leave the system.
Interactive Three.js model of a community water point helper system. Toggle assembly view. Drag to rotate, scroll to zoom. Stress slider changes sensor status and actor path.
How this model works: The full view shows a realistic community setting with a tank, solar panel, sensor nodes and a person approaching. When stress rises, the sensor changes colour and the actor re-routes toward an alternative point. In assembly view the parts separate so school children and grandparents can see how the solar power, sensor, communication device and water point work together. Materials use simple but realistic lighting and proportions. The model is for understanding only; it is not a live control surface.
Every dry borehole left invisible is another family walking kilometres or buying water. WaterPoint Integrity exists so the first signal of failure reaches the people who can act, in the language the community already uses, while a human still decides the response.
National non-revenue water (2025 No Drop progress). Early detection of point failures reduces downstream tanker dependence.
Annual value of water lost nationally. Faster response at community points protects both volume and public health.
Wastewater systems in critical condition (Green Drop 2025). Complementary integrity monitoring for drinking-water points is required.
Earlier alerts, multi-language messages, reduced time spent collecting unsafe water.
Prioritised, evidence-backed work orders instead of reactive tanker runs.
Shared status and coordination drafts that respect existing governance layers.
Low-cost solar nodes match existing borehole and standpipe infrastructure. Multi-language generative drafts address the documented gap between English municipal systems and community languages. Human-in-the-loop design aligns with POPIA and operational reality. Pilot deployments can start with 20–50 points per district without replacing bulk SCADA.
Plain-English context: Rural villages and informal settlements repeatedly lose access when communal boreholes fail, pumps are removed, or tankers cannot keep up. Health impacts follow. The sources below are specific to this geography and problem.
Problem: Communal boreholes stop after pump removal or transformer failure. How this solution improves it: Edge sensors flag the exact point and time; multi-language alert reaches ward committee and traditional leader within minutes. AI models: edge anomaly + GPT-class drafter.
Problem: Contaminated municipal water forces elderly residents to a hole. How: Visual confirmation model plus level sensor on temporary tanks; human reviewer prioritises safe alternative. AI models: vision model + coordination drafter.
Problem: Hundreds of millions spent on tankers while infrastructure fails. How: Integrity data shows which points are truly down so tanker runs are targeted. AI models: anomaly detector + human reviewer.
Problem: Municipal systems do not automatically share status with traditional leaders. How: Drafts generated in local languages and routed for dual review. AI models: multi-language generative model.
Problem: Stomach problems reported after borehole failure in Makhuvha. How: Faster restoration reduces exposure window. AI models: alert drafter with health-context notes.
Problem: Many rural points have no grid power. How: Solar node continues reporting during municipal power cuts. AI models: edge inference on solar gateway.
Problem: Reports of “no water” are hard to prioritise. How: Optional photo analysed by vision model before human review. AI models: GPT Image class candidates.
Problem: Vague complaints lead to wrong crews. How: Structured recommendation includes location, last known flow, and suggested skill set. AI models: coordination drafter + human approve.
Problem: English-only municipal messages are not understood by all residents. How: Drafts in Sepedi, isiZulu, Afrikaans generated on demand. AI models: Claude/Gemini multi-lingual candidates.
Problem: National systems cannot see every community point. How: Low-cost nodes and cloud aggregation give district-level visibility without replacing SCADA. AI models: edge + cloud ensemble under evaluation.
Exactly 20 areas (mix of provinces and cities). Live status uses real Open-Meteo weather fetch as an environmental stressor proxy; water-point telemetry is pilot demonstration data.
Select a location to see live environmental data flow into the decision-support view.
Source: Open-Meteo public API. Timestamp shown below. This demonstrates data arrival; actual water-point telemetry remains inside the pilot environment.
WaterPoint Integrity is a product of SEVEN15 PROJECTS.
Registered name: SEVEN15 PROJECTS
Registration number: 2026/880966/07
SARS income tax reference: 9200230291 (registered 3 November 2026)
Address: 391 Herman Street, Ratanang, Jacobsdal, Free State, 8710
Owner: Pacifist Nkonki
Email: pacifist@seven15projects.co.za
Water is the first thing a family needs each morning, and the first thing they worry about when a borehole or tap stops working. Too often the people who can fix the problem only hear about it days later, after children have walked far for water or families have paid for water they should not have to buy.
At SEVEN15 PROJECTS we work to close that gap. WaterPoint Integrity helps a failing water point be seen early, explained in the language the community speaks, and passed to the people who can act. The technology only drafts and suggests. A real person always makes the decision, because water is about people and dignity, not only pipes and sensors.
We are in the pilot stage and we are learning with municipalities, traditional authorities and communities. If you work with water in South Africa, I would like to hear from you.
Pacifist Nkonki
Owner, SEVEN15 PROJECTS
pacifist@seven15projects.co.za
Issued by the South African Revenue Service on 3 November 2026. Registered name: SEVEN15 PROJECTS. Taxpayer reference: 9200230291.
Download tax certificate (PDF)Direct: info@seven15projects.co.za · rfq@seven15projects.co.za
SEVEN15 PROJECTS, registration number 2026/880966/07, 391 Herman Street, Ratanang, Jacobsdal, Free State, 8710, South Africa. Contact: info@seven15projects.co.za.
A live testing product in pilot stage for the WaterPoint Integrity AI+IoT system. All telemetry, predictions and AI outputs shown in the demo workspace are part of the pilot demonstration environment.
Contact form collects name, email, organisation and message solely to reply to enquiries. Retention: 24 months or until deletion request. No marketing use.
Encryption in transit, access controls, data minimisation, retention limits, breach notification process, human-in-the-loop controls, and audit logging are design goals for any production deployment. The current pilot implementation has the security limitations of a demonstration environment.
Minimal or none. Session state for the demo is kept in memory only.
Open-Meteo for weather demonstration. Three.js from CDN. See their respective policies.
Access, correction, deletion, objection. Exercise by email to info@seven15projects.co.za.
POPIA for South African residents and operations. General data-protection principles apply for international visitors. Language is provided for international visitors while protecting the operating company.
This is a pilot-stage testing product. Humans always decide. No automatic approvals occur. Security limitations of the current implementation apply. The product generates decision-support drafts only.
Liability is limited to the maximum extent permitted by law. Users may not misuse the system to generate false alerts or harm communities. The company and its solutions are protected against claims arising from pilot demonstration use.