Problem

A university project that needed the shape of a real product: a learning platform with live sessions, users, content, and progress tracking — large enough that a single Spring Boot monolith would make ownership and deployment impossible for a student team.

Architecture

10 microservices

Spring Boot services split by domain, each with its own schema. No service reaches into another's tables.

API Gateway

Single entry point handling routing, authentication, and inter-service communication, so clients never learn service topology.

Kafka

Asynchronous messaging between services for the flows where a caller must not fail because a downstream service is slow.

Keycloak

Centralised authentication and authorisation across every service — tokens verified rather than credentials re-implemented per service.

LiveKit

Realtime video and group calling, integrated rather than self-hosted from scratch.

Docker + Kubernetes

Every service containerised, with deployment workflows automated locally and in cloud.

Decisions and trade-offs

  • Ten services for a student project is expensive. The honest trade: a monolith would have shipped faster. Splitting bought independent deploys and clear ownership inside a team, and cost me the operational overhead of a gateway, a broker, and a centralised IdP — which is exactly the load I wanted to learn to carry.
  • Keycloak over a hand-written auth service. Writing JWT issuing myself would have been more educational but less defensible; token issuance, refresh, and role mapping are the parts of a backend that hurt users when subtly wrong.
  • Kafka only for asynchronous paths. Synchronous request/response stayed REST between services where a caller genuinely needs the answer. Making everything event-driven would have added a broker’s failure modes to flows that don’t benefit.