TypeScript & Node.js
Our default for business applications and APIs. One language across the stack, a mature ecosystem, and a type system that makes refactoring safe.
- Fastify, Express, NestJS
- Strict mode, no implicit any
A deliberately boring technology stack
We keep to mainstream, well-documented tools. Not from caution, but because your system has to be maintainable by engineers who are not us — which is exactly the position you should be in.
01 — The rule
Four questions before anything enters a project
Every dependency is a commitment somebody has to keep for years. Before we add one, it has to survive these.
- Can we hire for it? If the local market cannot staff it, you are buying a dependency on us specifically.
- Will it exist in five years? Governance, release history and the size of the community matter more than benchmarks.
- What does it cost to remove? Anything that reaches into every file is a decision that cannot be reversed cheaply.
- Does it earn its weight? A library that replaces thirty lines of code is not a saving.
When a project genuinely calls for something unusual, we make the case in writing, note the trade-off in an architecture decision record, and get your agreement before committing.
This website, as a worked example
Twenty pages, no framework, no build step in production, no third-party requests, no cookies. The whole thing is HTML, one stylesheet, one script and two self-hosted typefaces.
- JavaScript shipped
- One file, ~13 KB before compression
- Third-party requests
- None
- Cookies
- None
- Runtime dependencies
- None
It is not the right architecture for every product. It is the right architecture for this one, and that distinction is the entire point.
02 — The stack
What we work with day to day
Not an exhaustive list of everything anyone here has ever touched — the things we are genuinely fluent in and willing to support.
Python
Where data processing, scripting or scientific libraries lead. Typed, linted and packaged like any other production code.
- FastAPI, Django
- pandas, pydantic
Go
For services where predictable latency, small binaries and low memory matter — integration workers, high-throughput APIs, CLI tooling.
- Standard library first
- Single-binary deployment
- TypeScript
- Node.js
- Python
- Go
- SQL
- Bash
Standards first
Semantic HTML and modern CSS do more than most teams assume. Grid, container queries, custom properties and view transitions replace a lot of former JavaScript.
- No runtime CSS-in-JS
- Design tokens in CSS variables
React & Next.js
When an interface has genuine application state. Server components where they reduce shipped JavaScript, client components where interaction requires them.
- React, Next.js, Vue
- Vite for tooling
Mobile
React Native for shared codebases; Swift and Kotlin when the app leans on the platform. The choice is argued per project, not assumed.
- React Native
- Swift, Kotlin
- HTML
- CSS
- React
- Next.js
- Vue
- React Native
- Swift
- Kotlin
PostgreSQL
Our default system of record. Constraints in the database rather than only in the application, and migrations versioned alongside the code.
- Foreign keys and checks enforced
- Reversible migrations
Caching & queues
Redis for caching, locks and lightweight queues. Dedicated brokers when delivery guarantees justify the extra moving part.
- Redis
- RabbitMQ, Kafka
Analytical stores
When reporting queries start hurting the transactional database, we separate them rather than adding indexes forever.
- ClickHouse
- Materialised views
- PostgreSQL
- MySQL
- Redis
- ClickHouse
- SQLite
Infrastructure as code
Terraform in your repository, reviewed like application code. No environment exists that cannot be rebuilt from a commit.
- Terraform
- Docker
Runtime
Containers on a managed platform by default. Kubernetes when the scale and the team genuinely justify the operational cost.
- ECS, Cloud Run
- Kubernetes
Observability
Structured logs, metrics and traces from the first deployment, not retrofitted after the first outage.
- Prometheus, Grafana
- OpenTelemetry
- Terraform
- Docker
- Kubernetes
- AWS
- Google Cloud
- Hetzner
- Grafana
Automated tests
Unit tests for logic, integration tests at every boundary, end-to-end tests for the journeys whose failure a customer would notice.
- Vitest, Jest, pytest
- Playwright
Static analysis
Type checking, linting and formatting enforced in CI so review conversations are about design rather than semicolons.
- ESLint, Ruff
- Prettier
Supply chain
Dependency scanning on a schedule, pinned versions, and an upgrade plan rather than an annual panic.
- Automated advisories
- Lockfiles committed
- Playwright
- Vitest
- pytest
- ESLint
- GitHub Actions
03 — A note on AI tooling
Where machine assistance fits, and where it does not
Our engineers use AI assistants the way they use any other tool — for boilerplate, for exploring an unfamiliar API, for a first draft of a test. What does not change is that a human reads every line, a second engineer reviews it, and the automated suite has to pass before anything merges.
We do not send client source code or production data to a third-party service without your written agreement, and where a client prohibits such tooling entirely we work without it. If you would like this stated explicitly in the contract, ask — we will put it there.
Already have a stack you have to live with?
Tell us what it is. We would rather work inside your constraints than sell you a rewrite you do not need.
HALDIR LTD · HE 459882 · Nicosia, Cyprus