Inspiration

Millions of day labourers on construction sites in Nigeria, Kenya and India are paid in cash from a paper register. Disputes are constant ("I was here all day" / "I completed my task"), workers leave with no record of what they were paid, and contractors can't prove what they paid if there's a complaint. Attendance apps exist, but few close the whole loop: proof of work, human approval, then a digital payout with a clear reason.

What it does

SitePay turns one site day into a paid, traceable record.

  • Worker on WhatsApp: taps Check in, shares their location and sends a photo from the site. No app to install. They're only checked in once both arrive. During the day they send work photos, and they can claim overtime.
  • AI check: the AI reads each photo and answers three questions for the supervisor: Present on site? Photos genuine? Work done? It flags late check-ins from the time stamped on the photo, reused photos from another day, and work that isn't finished. It gives a verdict ("AI confirms… safe to approve" or "AI flags this day: late check-in"), so the supervisor doesn't have to study every photo.
  • Supervisor dashboard: a register (Present, Absent, Half day, Overtime, advances). The supervisor approves or rejects each day. Rejected workers are told why on WhatsApp. The AI never approves or pays.
  • PayPal payout: one click pays a single worker or everyone. Only approved days are paid, and each payment says what it's for.
  • Pay note: when PayPal confirms, the worker gets a WhatsApp receipt: days, overtime, advances, net pay and the PayPal reference.
  • Scheduling: the supervisor plans who works which day, and the worker is told the date, start time and task on WhatsApp.

How PayPal, AG Grid and Bryntum power SitePay

PayPal: turning cash into a digital trail

Cash leaves no record. PayPal is what makes SitePay's pay provable.

  • Payouts API pays the crew straight from the contractor's business account to each worker's PayPal. A supervisor can pay one worker (e.g. someone who needs their money today) or everyone in one batch.
  • Every payment explains itself. Each payout item carries a note: "Lekki Phase 2 Duplex | 5–11 Nov | day labour. Present 5 days, Overtime 1 day. Gross ₦74,500, advances ₦0, net ₦74,500." Worker and contractor both have the same record if there's ever a dispute.
  • Nobody is paid twice. Each send uses a unique sender_batch_id and PayPal-Request-Id, and pay lines are claimed before the call, so double clicks and retries are safe.

  • Cross-border friendly. Wages are kept in naira; PayPal pays in USD at a rate stored on every payment, so the conversion is always auditable.

AG Grid: a register supervisors already understand

Contractors think in a daily register: one row per worker, P / A / H / OT, advances. AG Grid lets SitePay keep that familiar paper layout while making it fast and accurate.

  • Edit in place: click a cell to change status (a P/A/H/OT/L dropdown), overtime hours or an advance. The day's pay recalculates instantly.
  • The AI verdict lives in the grid. A custom cell shows each worker's AI check (✅ verified / ⚠️ flagged, with the reason), so a supervisor scanning 30 workers spots the problem rows at a glance.
  • Approve or reject from the row. Custom cell buttons let the supervisor approve or reject a day without leaving the register. Unapproved WhatsApp check-ins show as P? until confirmed.
  • Bulk actions: select rows to mark several workers Present, Half day or Absent, or approve them together.
  • Pinned worker column and computed "Day pay" column keep the register readable on a busy site day.

Bryntum Scheduler: planning the crew ahead

A register only looks back; a site also needs to plan. Bryntum Scheduler gives SitePay a two-week crew calendar.

  • One row per worker, one block per day, coloured by what actually happened (Present, Overtime, Half day, Absent). Days already paid are faded, so the supervisor sees the whole fortnight at a glance.
  • Schedule future shifts: double-click an empty future day (or use Schedule worker) to plan a shift with dates, start time and task. The worker immediately gets it on WhatsApp: "📅 You've been scheduled at Lekki Phase 2 Duplex, Thu 12 Nov, start 08:00, task: blockwork, 2nd floor."
  • Planned shifts look different (striped) from worked days, and clicking one cancels it, which also tells the worker on WhatsApp.
  • Hover tooltips show the day's status, overtime, advances, photo count and review state; a live time line marks "now".

How we built it

  • Backend: Python + FastAPI, MongoDB Atlas, running in Docker.
  • WhatsApp Cloud API: webhook for check-ins, photos and location; outbound schedule messages, rejection notices and pay notes.
  • AI: a vision model (Gemma, through an OpenAI-compatible endpoint) reads check-in and work photos and returns structured findings. Deterministic checks run alongside it: a perceptual hash catches reused photos, brightness catches dark photos, and the photo time stamp catches late arrival.
  • PayPal: OAuth client credentials plus the Payouts API (POST /v1/payments/payouts, GET /v1/payments/payouts/{id}), in sandbox.
  • Dashboard: React + Vite, AG Grid for the register, Bryntum Scheduler for the site calendar, hosted on Firebase.
  • Photos are re-encoded on arrival to strip GPS and other metadata before they're stored.

Challenges we ran into

  • Making the AI produce a verdict, not a description. "A mason taking a selfie" doesn't help a supervisor; "Present, but late: photo stamped 12:00" does.
  • Paying only once: per-worker and pay-all buttons both had to be safe against double clicks and retries.
  • WhatsApp's 24-hour messaging window, which limits when a pay note or schedule message can be delivered.
  • PayPal can't pay out in naira, so wages convert to USD with the rate stored on every payment.

Accomplishments that we're proud of

  • The full loop works end to end in sandbox: WhatsApp check-in, AI verdict, supervisor approval, PayPal payout.
  • The supervisor stays in control: AI flags, human decides, PayPal pays.
  • Each sponsor tool solves a real part of the problem: AG Grid for the daily register, Bryntum for planning the crew, PayPal for the paper trail.

What we learned

AI is most useful here as a careful assistant that structures evidence, not as a decision-maker. The trust in the system comes from the human approval step and the PayPal record.

What's next for SitePay

  • Pay into local wallets and mobile money where workers don't have PayPal.
  • Voice-note check-ins for workers who can't read.
  • Add other Nigerian languages for loborers that can not speak english
  • Multi-site contractors and weekly pay schedules on the Bryntum calendar.
  • A worker history workers can show as proof of income.

Built With

Share this project:

Updates

Submission history