Case study
SkyBiz Payouts
Payments back office with risk review
1 payout per withdrawalin tests, even with retried jobs and webhooksat-least-once delivery with idempotent consumers (single effect)
Live demo
No public demo yet: it is an internal back office.
What it is and what it solves
A payments back office: deposits through Stripe (test mode) and a crypto simulator, withdrawals with risk scoring and manual review, reconciliation and AI-assisted promotions (multilingual drafts approved by an operator). It moves money in and out of Platform Core with reliable jobs and an auditable event trail.
Business version: SkyBiz →
Architecture
- Back office → Payouts API
- Payouts API → RabbitMQ (jobs)
- RabbitMQ → Workers
- Workers → Stripe (test)
- Workers → Platform Core (deposit · withdraw)
- Workers → PostgreSQL
- PostgreSQL → Kafka (outbox)
Simplified from the project description; the repository is private.
Three technical decisions
RabbitMQ for jobs, Kafka for events
Jobs go to RabbitMQ (TTL retries, dead-letter queue, VIP priority, manual ack) with at-least-once delivery and idempotent consumers, so each request has a single effect. Facts other services subscribe to leave through a transactional outbox to Kafka.
Risk scoring with a human in the loop
Withdrawals are scored and the risky ones wait for manual review, so automation speeds up the normal case without approving the dangerous one.
Reconciliation as a feature
Provider records and the ledger are compared on purpose, so any mismatch is surfaced instead of discovered later.
Tests, with numbers
- 57 tests against real PostgreSQL and RabbitMQ.
- 7 faults injected on purpose, all detected — for example a worker killed mid-job, a message that exhausted its retries and a duplicated webhook.
Verified
- Integrated with Platform Core.
Not claimed
- Stripe runs in test mode; crypto is simulated.
- Figures come from test runs, not from production traffic.
Stack
Code and CI
Private repository · code walkthrough on request