Make sure the right person gets the right support, and fix it fast when they don't.
Solar-powered verification points read card taps, consent-based photos and GPS time-stamps at pay points and distribution days. The system compares them with the official list, flags mismatches, drafts plain-language alerts in all 11 spoken official languages, and puts the evidence in front of a named human officer to decide.
Pilot test data
Pilot test data
Pilot test data
Support fails quietly: at the list, at the queue, and in the complaint that goes nowhere.
The eight steps below show how one weak point in a beneficiary list becomes a lost grant, a disputed house or a missed food parcel, and how trust wears away until the cycle starts again.
Three kinds of verification failure
Inclusion errors
People who are not entitled receive support: ghost, duplicate or deceased names, or insiders using their access. Reported cases in the IEJ analysis involve government officials far more often than beneficiaries.
Exclusion errors
People who are entitled are left out: names dropped or never added, no ID document, no phone, no data, or a review they cannot reach. Elderly and disabled people queue for days to prove what the system already should know.
Exploitation
Third parties and insiders profit from beneficiaries: unlawful deductions, sold places on housing lists, misuse of personal data. The Constitutional Court's Black Sash litigation made data protection part of the grants story.
From a card tap to an approved fix in seven steps.
Scenario Theatre: five different days, one engine
Pick a scenario, press Play, and watch the pipeline run. Every scenario pauses at the point where a person must decide. Numbers are computed by the run itself and are pilot test data, never field results.
Results of the runs you completed
| Scenario | Trigger | Stage that caught it | Human decision | Outcome | Time to decision (this run) |
|---|
Rows appear as you complete each scenario. Pilot test data
Eight AI roles and six cloud roles, each with a human in control.
Model and service names were checked against provider documentation on 7 November 2026; each card links to the provider's documentation. Technologies are named factually and no provider is described as a partner or endorser.
AI Helpers
Cloud Helpers
How the helpers work together
How they talk to each other
How they talk to each other (3D)
drag-free · auto animatedSee the whole solution at a pay point, running.
Drag to turn, scroll or pinch to zoom, click any part to learn what it does. Use the slider to see what happens when more people cannot be matched to the list.
Calm day: nearly everyone matches the list.
How this model works
The pole at the pay point carries a solar panel, a power box with a battery, and the sensors: a camera, a card reader and a GPS puck. When a person taps their card, the tablet checks the name against the official list for that day. If many names cannot be matched, the beacon changes colour, the tablet shows a warning, the queue grows, and an officer's phone gets the evidence. Nothing is approved by a machine: the officer decides what happens next.
Legend
Green: calm, names match.
Amber: attention, more names than usual cannot be matched.
Red: high risk, a human check is needed now.
Drifting paper sheets show disputed lists. The blue figures are matched people; amber figures could not be matched.
What changes when lists, taps and photos can be compared the same day.
A note from the director
Assessment
Who is helped
Pensioners, people with disabilities and caregivers who today queue for days or lose payments after a review they cannot reach. A verification point comes to the pay point or hall instead of asking people to travel to a portal.
Who is protected
Officials and community leaders who must choose between households with little evidence. Each decision is recorded with its evidence and the name of the person who made it.
What is not claimed
No saving, accuracy or fraud-reduction figure is claimed for the field. Figures on this site are pilot test data from a scripted simulation and are labelled that way.
Stakeholders
Municipal and provincial officers
See flags, evidence and recommended actions in one place, with a printable summary for each pay day.
Traditional authorities and ward committees
Receive plain-language summaries and can record vouching for a person without a document. Vouching is recorded as such, never as proof.
Beneficiaries
Are told in their own language what was checked and what to do next. Consent is requested for photos; refusal never removes support by itself.
Field technicians
Receive work orders for failed sensors, flat batteries and weak signal, with a checklist and a verification step.
Research and development
Focus areas: fairness testing across languages, ages and skin tones; offline-first behaviour under load-shedding; consent wording tested with community members; independent review of the audit trail design; data protection impact assessment; and support for sign language and low-literacy users.
Ten verified sources on why beneficiary lists fail.
Each source was opened and read on 7 November 2026. Summaries are paraphrased. Press coverage and research are labelled as such; none of them is an endorsement of this solution.
Validation panel
| Check | Result |
|---|---|
| Number of sources | 10 |
| Official, regulator or court sources | E4 (official), E6 (regulator), E9 (court proceedings) |
| Research and first-hand accounts | E1, E7, E8 |
| International body | E10 (World Bank) |
| Date accessed | 7 November 2026 |
Ten problems, ten articles, one to one.
Each problem below is a short article tied to one verified source from the Evidence page.
Twenty areas: nine provinces and eleven towns and cities.
Weather is fetched live from Open-Meteo when you open this page. "Live" on this site means a real network fetch; if it fails the card says so.
Who stands behind this solution.
Company details
Qinisekile is a product by .
| Enterprise name | |
| Enterprise type | Private Company |
| Registration number | 2026/873298/07 |
| Registered with | Companies and Intellectual Property Commission (CIPC), South Africa |
| Registration date | 6 November 2026 |
| Business start date | 6 November 2026 |
| Status | In Business |
| Financial year end | October |
| Registered office | |
| Postal address | |
| Director | Noluthando Michelle Ngwenyama |
| Managing Director | Edwin Michael Chala |
| info@auravyn.co.za | |
| Phone | |
| Website | auravyn.co.za |
Mission
Help the right person get the right support, and make it easy for a human to correct mistakes quickly, in the person's own language.
Principles
A human always decides. Consent for photos. Minimal data. A visible audit trail. No claim of endorsement by any programme, department or provider.
Leadership
Talk to us, or raise a concern.
Directory
Email:
Phone:
Address:
Information Officer:
Email addresses
Information Regulator (South Africa)
enquiries@inforegulator.org.za · 010 023 5200 · toll-free 0800 017 160
Woodmead North Office Park, 54 Maxwell Drive, Woodmead, Johannesburg, 2191
POPIA complaints: POPIAComplaints@inforegulator.org.za
inforegulator.org.za/contact-us
SASSA
SASSA online services portal. This site is not affiliated with SASSA.