David Soldat #00Software engineer, Banja Luka

Green flag: open to backend and full-stack roles

Back to the pit wall
1BRPFastest lap

BridgePay

A Klarna-style "Pay in 4" platform. A trained credit-risk model decides in real time whether to extend credit, the merchant is paid in full, and the platform collects the installments afterward.

Try the live demoView code on GitHub

Setup sheet

How the car is built, part by part
Power unit
Java 21 and Spring Boot 4
Chassis
7 independently deployable microservices behind Spring Cloud Gateway
Bodywork
Three Angular 21 apps with standalone components, signals and Tailwind
Fuel system
PostgreSQL with a schema per service, Flyway migrations and Redis
Radio
Kafka in KRaft mode, transactional outbox, idempotent consumers
Strategy
Logistic regression trained in scikit-learn, served through ONNX Runtime, AUC-ROC 0.82
Safety
Keycloak sign-in, and a circuit breaker on every external call
Garage
Docker images built by GitHub Actions, pushed to GHCR, running on k3s (ARM64)

Car layout

Frontend appServiceML or AI modelData storeExternal
StorefrontPay in 4 checkoutMain appOps and merchantsKeycloakOIDC with PKCEAPI GatewayRouting, JWT, rate limitsApplicant serviceSignup and identityApplication serviceCheckout and decisionsKafka event busRepayment reconciliationInstallment plansPaddlePaymentsMock credit bureauKaggle dataset rowsCredit risk engineONNX model, RedisNotificationsIdempotent consumersign inscoreoutboxinstallmentswebhooks
  • Storefront to API Gateway
  • Main app to API Gateway
  • Main app to Keycloak, sign in
  • API Gateway to Applicant service
  • API Gateway to Application service
  • API Gateway to Repayment reconciliation
  • Application service to Credit risk engine, score
  • Credit risk engine to Mock credit bureau
  • Application service to Kafka event bus, outbox
  • Kafka event bus to Repayment reconciliation
  • Repayment reconciliation to Kafka event bus
  • Kafka event bus to Notifications
  • Repayment reconciliation to Paddle, installments
  • Paddle to Repayment reconciliation, webhooks

Lap breakdown

  1. Sector 1, Architecture

    Seven services, one gateway

    Each service deploys on its own and owns its PostgreSQL schema. Spring Cloud Gateway handles routing, JWT validation, rate limiting and correlation IDs, so the services stay focused on the business flow.

  2. Sector 2, Data and AI

    A model that explains itself

    I trained a logistic regression on bureau-style features (AUC-ROC 0.82) and serve it through ONNX Runtime. Every decision stores per-feature contributions, which the ops review screen shows as an explanation chart.

  3. Sector 3, Delivery and integrations

    Fails safe, never blind

    Decisions and repayments travel over Kafka through a transactional outbox. Every external call has a circuit breaker, and any scoring failure sends the application to manual review. Paddle webhooks are verified, and failed events land in a retry queue.

Onboard footage

Screenshots from the app
Ops dashboard: decision mix and score distribution
Storefront checkout
Merchant sales and checkout funnel