Articles & case studies

How production problems actually got solved.

Real systems, real numbers, client names removed. Each piece walks through the problem, the options measured, and what shipped.

Architecture

Designing and shipping a two-sided marketplace as its only engineer

MMA Finds connects fighters, managers, promotions and their staff. The architecture behind a production marketplace on iOS and Android: a core backend with explicit domains, a Kafka-driven notification service, payments that cannot double-credit, and an analytics dashboard for the business.Java 17 / 24Spring Boot 3PostgreSQLKafkaKeycloakStripeExpo / FCMReact NativeReactDockerNginxHetznerRead →Reliability

A job that can crash, retry, or run twice — and still never lose an order

Publishing a dataset meant generating several 300,000-record export files inside a web request. The asynchronous pipeline that replaced it: leased claims with SKIP LOCKED, automatic recovery, duplicate rejection at the database, self-cleaning failures and once-only catch-up after downtime.Spring BootPostgreSQLRDF4JKubernetesRead →Observability

From "grep every pod and hope" to one trace ID across a 7-service chain

A 7-service data pipeline with a RabbitMQ hop, two NiFi hops and a callback had no correlation between logs, traces and metrics. The observability stack built from zero: OpenTelemetry Collector, Grafana, Tempo, Loki, Prometheus, wired once in a shared library.OpenTelemetry CollectorGrafanaTempoLokiPrometheusSpring BootKubernetesRead →Performance

Cut filtered dataset search from 30+ minutes to sub-second at 25–50M records

A data-grid search over tens of millions of MongoDB documents took 30+ minutes in the reported worst case. Four stacked query-design problems, fixed in stages and benchmarked before release, with no new infrastructure and no new license.MongoDBSpring BootKubernetesAngular / PrimeNGOpenSearch (evaluated)Read →