Skip to content
All systems
LiveFlagshipKSDRILL-SA · Founder & Principal Engineer / Fintech

Xkimi Xa Mali

A private savings collective platform, built for real money and real family stakes.

  • Next.js
  • React
  • TypeScript
  • PostgreSQL
  • Prisma
  • Inngest
  • Redis
  • Vercel+4
View it liveView the source
Started
May 2026
Shipped
October 2026
Last push
4 days ago
Commits this year
537
xkimixamali.co.za
Xkimi Xa Mali homepage

Written by AI from the repository · updated 7 weeks agoAI

The problem

A family savings group collects a monthly contribution from every member. Done over chat messages and memory, nobody can say with certainty who has paid, what the pool holds, or what happened to a payment that was later reversed. When the money is real, "roughly right" is not good enough.

How it works

Three Next.js applications share one backend: a member portal, an admin dashboard and a public website. The member portal also hosts the REST API (/api/v1, route → service → repository) and the scheduled jobs.

  • Ledger — every rand lands in an append-only, double-entry pool ledger in PostgreSQL. The balance is the sum of credits minus debits, and a nightly reconciliation job checks it.
  • Payments — an administrator records each transfer or cash payment against the member and the month, with proof of payment. A payment can be reversed but never erased.
  • Jobs — 23 durable, retryable Inngest functions: contribution reminders, overdue sweeps, month rollover, ledger reconciliation, financial-anomaly watch, backups and more.
  • Members — each member has a live view of their standing, a year-end forecast and on-time rate, group goals with pledges, and an inbox fed by SMS, email and in-app notifications.

Engineering decisions

The repository records its decisions as architecture decision records:

  • PostgreSQL on Neon, chosen over Supabase — ACID transactions are non-negotiable for money.
  • Inngest over plain cron — jobs that touch money must be durable and retryable.
  • Netcash DebiCheck over PayFast — a South African recurring-debit mandate. The debit-order machinery is built and tested but deliberately dormant: a deployment with no collections provider refuses every money operation rather than pretending.
  • Encryption at rest — bank and ID numbers are AES-256 encrypted, with a documented key-rotation procedure.

Built to be trusted

  • 179 test files across the three apps.
  • 55 database migrations, applied by CI.
  • Six GitHub workflows: CI, governance, scheduled backups, a backup self-test, restore drills and preview-database clean-up.
  • POPIA-aware data requests and erasure, an audit trail, and admin-signed PDF statements.

Where it stands

All three apps are live and the operating model works end to end: members pay, an administrator records it with proof. Development paused at a deliberate point (September 2026) — what the group needs next is a record of real contributions that a collections partner will accept, and only time and real payments produce that.

Activity

How it's moving.

26 weeks agothis week
commits in the last 26 weeks
537

Latest commits

  1. fix(reports): give PDF export room to finish, and a real error message on failure
  2. fix(admin): surface stuck-PENDING members and let an admin resend verification
  3. fix(utils): invite form default matches the actual minimum
  4. fix(admin): every invitation was failing server-side, plus a real misclick bug
  5. chore(web): remove the fee-buffer retro audit, now that it has run
  6. chore(web): add a read-only audit for fee-buffer retro candidates
  7. fix(web): carve out bank-charge padding on an offline payment, not overpayment
  8. fix(web): enforce June 2026 as the earliest contribution period, and due dates fall at month-end everywhere

Can I measure visits?

Anonymous page-view counts (Vercel Analytics, no cookies) help me see what’s useful. Nothing is measured unless you say yes, and you can change this any time from the footer.