Every rand paid. Every day worked. Every skill recorded.
ProofWork, a product of Meridian Advisory
ProofWork checks young workers in with a QR code or biometric device, confirms their location with GPS, and turns verified attendance into skills-transfer and outcome reports. It flags things that look wrong. A programme officer decides what to do.
All charts on this page are illustrative and simulated. They are not real programme data.
This solution requires current AI models to operate.
Why verification matters
Public works programmes employ many young people at scattered sites. Paper registers make it hard to confirm who worked, where and how well. That hurts honest workers most, because their results are hard to prove.
Verified
QR or biometric check-in matches a GPS location inside the site fence.Needs a look
Duplicate scan, location mismatch or work photo that does not match the task.Not enough evidence
Device offline or data missing. The officer is asked to confirm on site.How it works
1 Check in
QR badge or biometric device at the site.2 Confirm location
GPS check against the site boundary.3 Validate attendance
Patterns checked for duplicates and impossible movements.4 Confirm work quality
Site photo compared with the assigned task.5 Report skills
Skills-transfer report for each young worker.6 Evaluate outcomes
Programme outcome evaluation for funders.7 Officer decides
Verify, ask supervisor, escalate or clear.The AI helpers
ProofWork uses four AI helpers, each with one narrow job, and one human who is always in charge. The helpers read the evidence and prepare drafts and flags. The programme officer reviews every flag and makes every decision. Here is how the whole thing works, from a worker arriving at a site to a report reaching a funder.
GPT-6 Luna: attendance validator
Checks every check-in and spots patterns that look wrong.
- Reads: QR or biometric check-in, time, GPS position, site fence, device status.
- Looks for: the same badge at two sites, check-ins outside the fence, movements that are impossible, devices that went quiet.
- Produces: a verified, needs-a-look or not-enough-evidence label, with the reason in plain words.
- Never does: accuse a worker or stop a payment.
GPT-Image-2.5 (Sunburst): work quality checker
Compares site photos with the task that was assigned.
- Reads: a supervisor's site photo and the task description, for example kerb repair in Ward 12.
- Looks for: whether the photo plausibly shows that task, and whether it looks reused or taken somewhere else.
- Produces: a match or mismatch note attached to the day's record.
- Never does: judge a worker's effort. A mismatch only asks a person to look.
GPT-6.1 Sol: skills-transfer reports
Turns verified days of work into a record the young worker can carry to their next job.
- Reads: only verified attendance and the tasks done on those days.
- Produces: a plain-language report of skills practised, hours worked and suggested next steps, such as a short course.
- Checked by: the programme officer before anything is shared with the worker or a future employer.
GPT-6 Astra: outcome evaluations
Summarises how a whole programme cycle performed, for municipalities and funders.
- Reads: verified attendance rates, flag outcomes and skills data across sites.
- Produces: an evaluation with what worked, where coverage was weak and what to change next cycle.
- Checked by: the programme officer and, where needed, the internal auditor and evaluator.
Programme officer
Reviews every flag. Nothing changes for a worker without them.
- Sees: the flag, the evidence behind it and the helper's reason.
- Chooses: verify on site, ask the supervisor to confirm, escalate to the internal auditor, or clear the flag.
- Recorded: every decision is logged with the time, so it can be audited and appealed.
One check-in, step by step
1 Arrive
A young worker scans a QR badge or uses a biometric device at the site.2 Locate
The device records the time and GPS position and compares it with the site fence.3 Validate
Luna checks the check-in against earlier ones and against other sites. Most check-ins pass and need no attention.4 Photo
The supervisor uploads a site photo. GPT-Image-2.5 compares it with the assigned task.5 Flag
If something looks off, a flag is created with the reason. If data is missing, the flag asks for an on-site check instead.6 Decide
The programme officer reviews the flag and chooses an action. A verifier is sent out only after the officer approves.7 Report
Sol builds each worker's skills report from verified days. Astra builds the programme evaluation for the funder.Safeguards built in
Human decides
AI only raises flags and drafts reports. People approve every action.Leads, not findings
A flag means "please look", never "fraud". Workers are not treated as guilty.Consent and POPIA
Biometric and location data is collected only with the worker's consent and handled under POPIA. A POPIA impact assessment is completed before any pilot.Bias and false flags
Before each pilot, flags are tested for bias and for false alarms, and thresholds are adjusted with the programme team.Appeal route
Every worker has a clear way to question a flag or a record, and the officer's decision is logged.Works offline
Where a device or the network is down, the record is marked "not enough evidence" and the officer confirms on site.AI dependency. This solution requires current AI models to operate; the models listed (or their current equivalents) are essential to the production system.
Connected city model
How this model works
The model shows a small town with ProofWork check-in devices at work sites. Each glowing node is a device that records who arrived and where, and sends it to the cloud shape above the town. Use the Assembly view to pull the layers apart: the ground and buildings, the devices, and the cloud that checks attendance. The anomaly slider shows what happens when check-ins start to look wrong: nodes turn amber, then red. Above 70 the verifier vehicle drives to a flagged site, but only a programme officer decides what happens next.
Drag to rotate, scroll or pinch to zoom. Each glowing node is a check-in device at a work site, linked to the central cloud shape. As the slider rises, nodes turn amber then red as anomalies appear. Above 70 a verifier heads out to a flagged site. Simplified for teaching; no real data is shown.
Context and further reading
Background on South Africa's Expanded Public Works Programme (EPWP) and the public employment context ProofWork is built for.
Where we could start
Live weather from Open-Meteo, refreshed every 10 minutes. Rain and heat affect outdoor work. Locations are metros where we hope to pilot first; they are our own planning designations, not agreements.
About and team
ProofWork, a product of Meridian Advisory
Meridian Advisory is a South African start-up, founded in 2026 and based in Krugersdorp, Gauteng. ProofWork is our first product: a verification and reporting platform for youth public employment programmes. It confirms who worked, where and when, and turns that attendance into skills and outcome reports.
Where we are. We are an early-stage company. The platform on this site is a working demonstration built on simulated data, so every chart, score and site you see here is illustrative. We are building in the open and learning from the people who run and take part in these programmes.
How we work. We are a small, focused team. We build for programme officers, municipal coordinators, auditors and site supervisors, and we keep a person in charge of every decision. ProofWork flags what looks wrong and an officer decides what to do.
What we are looking for. We want to work with municipalities, implementing agents and funders who are ready to pilot verified attendance and skills reporting. To start a conversation, write to admin@meridianadvisory2428.co.za, or send tenders and quotation requests to rfq@meridianadvisory2428.co.za.
Registration
Entity: Meridian AdvisoryRegistration no.: 2026/780115/07
Tax status: Registered with SARS (Income Tax) · 30 September 2026
Registered address: 17 Seehoff Avenue, Mindalore, Krugersdorp, Gauteng, 1739, South Africa
ProofWork is a product of Meridian Advisory. 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 ProofWork 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, Meridian Advisory can deliver a production system that depends on current AI models and that those programmes are designed to help scale.
Anele Dlamini
Director · anele@meridianadvisory2428.co.za"I started Meridian Advisory because too many young people work hard in public programmes and still cannot prove it. When attendance and skills are verified, their effort becomes a record they can carry into their next job, and honest work is finally recognised. ProofWork gives programme managers trusted data and gives young people evidence of what they have done and learned. As it grows, I want it to open real doors for the youth of South Africa."
Why this matters
"Too many young people work hard in public programmes and still cannot prove it. When attendance and skills are verified, their effort becomes a record they can carry into their next job. ProofWork is how I want honest work to be recognised."
Anele Dlamini, Director, Meridian AdvisoryYoung workers
A verified record of days worked and skills practised, to show future employers and training providers.Programme officers
A short list of flags to review instead of paper registers to check, with the evidence in one place.Municipal coordinators
Clearer reporting on who worked where, and earlier warning when a site has problems.Funders and auditors
Attendance and outcome data that can be traced and audited, with every officer decision logged.Communities
Public money tied to work that was actually done, in the wards where it is needed.Articles
Ten short explainers on the problems ProofWork addresses. The AI models listed are candidates under evaluation, not endorsements.
Contact us
Legal, privacy and notice
Who we are
Meridian Advisory, Reg. no. 2026/780115/07, 17 Seehoff Avenue, Mindalore, Krugersdorp, Gauteng, 1739, South Africa. Email: contact@meridianadvisory2428.co.za.What this website is
A prototype and demonstration of ProofWork. Data, people, sites, scores and AI outputs in the demo are simulated, unless labelled live. The only live data is the weather on the Coverage page.Personal information
The Contact form collects the details you type (name, email, optional phone, message and, for business enquiries, your organisation) only to reply to your enquiry. The form opens your email app, so this site stores nothing. Your email is then held in our mailbox only as long as needed to respond and keep a record of the conversation.Cookies and local storage
This site sets no cookies and does not use local storage.Third-party services
Google Fonts (privacy), the Three.js library from cdnjs/Cloudflare (privacy) and Open-Meteo weather data (terms). These services may receive your IP address when the page loads.Your rights (POPIA)
Under the Protection of Personal Information Act (POPIA) you may ask to access, correct or delete your personal information. Write to contact@meridianadvisory2428.co.za. You may also contact the Information Regulator at inforegulator.org.za.Sign in
Demo access only. No account is created or verified.