What it is

A monorepo project: Java Spring Boot for the core services, PostgreSQL for storage, Kafka for event flow, Python/FastAPI for fraud detection, Next.js for the front end, Docker and Kubernetes for deployment, and Prometheus, Grafana, and OpenTelemetry for observability.

What it’s really about

The application layer is invoice financing, but the actual subject is correctness under failure: guaranteeing a settlement executes exactly once on top of a network that only guarantees at-least-once delivery, with every settlement carrying an explicit state (PENDING, CONFIRMED, FAILED, UNKNOWN, REVERSED) and a reconciliation process that catches drift against external reality instead of assuming success.

Where it stands

All nine planned build phases are done, deployed, and verified end to end on a live staging server. Load and chaos testing in the final phase found and fixed two real gaps in gateway failure handling and deadlock-retry exhaustion, alongside earlier concurrency bugs caught by integration tests before they ever reached a real environment.

What I’m doing with it now

Not building further, restudying. Having built it doesn’t guarantee being able to explain it cold, under interview pressure, and that gap is worth closing deliberately rather than assuming it isn’t there. The first subsystem studied, idempotency, surfaced a real crash-recovery gap, fixed the same day. Outbox is next.

What I took from it

  1. Explicit states beat assumed success. UNKNOWN is a real answer.
  2. Load and chaos tests find bugs design review doesn’t.
  3. Building something and being able to explain it cold are two different achievements.