01 Languages
Our centre of gravity is Kotlin and TypeScript, Python for data and systems work, C/C++ where performance or hardware is involved.
technology
What we do
Not a buzzword list from a sales deck — everything below runs in production, on our own systems and on our clients’. Where there is a choice, we pick the boring, maintainable one.
02 Backend and APIs
Quarkus + Kotlin for larger applications, Express/TypeScript for smaller services. Contract-first interfaces (OpenAPI), with GraphQL or gRPC where they fit. Schedulers, database-backed retry queues that survive a restart, templated mail, integrations with payment gateways and calendars.
03 Service architecture
We do monoliths and platforms of dozens of services alike. Asynchronous messaging over Kafka, secrets in Vault, single sign-on through Keycloak, declarative GitOps deployment.
04 Frontend
Vue 3 for admin interfaces and applications, Astro for static sites (no PHP, no database under a brochure). We also build standalone embeddable widgets that attach to a third-party page with a single <script> tag.
05 Data
PostgreSQL by default, with the schema versioned through Flyway migrations — never by hand on production. We have done in-place major upgrades of a live database and connection pooling when the pool runs dry.
06 Operations and infrastructure
Everything runs in containers, ships from a pipeline and keeps its configuration in git. From a single Compose file to a Kubernetes cluster behind load balancers. Our own GitLab, our own registry, our own certificate authority — so we can also work in environments that must not reach the cloud.
07 Monitoring, quality and security
Metrics, dashboards and alerts ship with the application, not after the first outage. Static analysis for code quality and error reporting from production. Our mail path is a full in-house chain including anti-spam and anti-virus.
08 Images, video and documents
Batch media processing: conversions and derivatives, generated web galleries, splitting 3D video, aligning an image series around a rotating object, typesetting documents to PDF.
09 Machine learning and LLMs
We deploy models where there is something to measure. Text classification with a locally hosted LLM (nothing leaves the building), predictions over time series, and above all backtesting — the result has to hold up on data, not in a presentation.
How we work
- Git is the source of truth. Configuration, migrations, content and infrastructure. An admin panel is a convenience, not an archive.
- The pipeline deploys, not a person. Merge to the main branch → build → deploy. The same procedure on the first release and on the hundred and fiftieth.
- Migrations are appended, never rewritten. An applied schema change is history, not a draft.
- We measure before we guess. Metrics and alerts are part of the delivery, not a separate order a year later.
- One build, many brands. Whatever is client-specific goes in configuration — not in a copy of the whole project.
- Documentation next to the code. So the next developer — including yours — has something to reach for.
Don’t see what you need? Most projects start with “is this even possible?” anyway. Write to us — you will get an answer even if it sends you elsewhere.