Reference Projects
Three projects from three industries — united by the same standard: clean messaging architecture, clear arc42 structure and demonstrable impact in production.
predigt.io is a fully self-developed SaaS platform for German-speaking congregations — realised from product idea to production operation under sole responsibility. The platform transcribes, translates and analyses sermons automatically in eleven languages and makes them semantically searchable.
At its core is an AI pipeline of three transcription engines: AssemblyAI for express results, OpenAI Whisper on a dedicated GPU for maximum precision and Claude (Anthropic) for summaries, chapter markers and contextual evaluations. All three engines run in parallel and are routed cost-aware.
Communication between eleven independent microservices is handled exclusively via NATS JetStream — with Object Store for audio transport, KV Store for distributed pipeline state and strictly typed Protobuf schemas as service contract. No synchronous RPC, no direct service-to-service dependency.
The operation runs on a self-built Hetzner Kubernetes cluster with a full observability stack, GitLab CI/CD and Helm-based deployments.
For a major operator of critical energy infrastructure, a target communication architecture for time-critical balancing energy processes was developed in collaboration with several partner companies. The project is subject to the BSI KRITIS regulation — every communication path must demonstrate an availability of 99.96 % across the entire process chain.
The core task was the systematic analysis of existing communication paths between European market platforms and the operator's backend systems, the evaluation of alternatives and the development of a consolidated architecture proposal according to arc42 — ready for decision by the relevant technical and IT committees.
The derivation took into account the procedural framework conditions of time-critical balancing energy processes, financial and time-based evaluation of all variants, and scenarios for a possible cloud migration of central system components.
To extend a core banking platform (Avaloq), a Kafka-based integration layer was designed and implemented. The goal was to decouple the grown monolithic interfaces through a reliable, scalable event streaming backbone — without interrupting ongoing banking operations.
Spring Boot microservices on OpenShift handle the domain processing. Connecting the Avaloq core system, readiness probes, default timeout analysis and performance measurements under load were also part of the task, as was advising the teams on Kafka concepts: partition sizing, event structure and versioning, linger.ms/batch.size tuning and cluster failover.
For mobile apps (Zak), REST interfaces were designed and implemented. An ELK stack enables end-to-end latency analysis across the entire Kafka pipeline.
