IT consultancy · Enschede, Netherlands

Built to hand over.

We design, build and run AI systems, data platforms and the web applications on top of them — on your infrastructure, by the people you'll actually talk to. The handover isn't a phase at the end. It's how we build.

Scroll

What we do

Four things, and how they fit together

Most of our work sits where these overlap: a model that needs an interface, an application that needs somewhere to run, a decision that needs evidence.

Data & AI engineering

Pipelines that get your data to the point where it can be asked questions, retrieval and tool-using agents on top of it, and the evaluation harness that tells you whether the whole thing works. We build for the day the model is wrong, not only for the demo where it is right.

Pythonpydantic-aiLLM agentsRAGMCPevalsdata pipelinesPostgreSQL

Custom software & web apps

From a bookkeeping tool that gives back an afternoon a week to a multi-tenant platform holding patient records. One team owns it from the browser tab to the database backup, so nothing falls between suppliers.

TypeScriptSvelteKitReactFastAPIPostgreSQLmulti-tenantSSO
In production: VoetLink

Platform & infrastructure

European hosting, described in code. Infrastructure you could rebuild from an empty server in an afternoon, with backups that have actually been restored.

infrastructure-as-codeOpenTofupyinfraDockerCaddyGitHub ActionsHetzner

Technology strategy

Technology radars, architecture reviews, and an honest answer on where AI pays for itself and where it does not. We advise on things we have built ourselves.

Technology radarArchitecture reviewbuild-vs-buyAI readiness
In production: Nodus

Selected work

Forward deployed

Our engineers work inside your team, on your problem, with your data. We sit with the people who do the job today, ship a first version they can use within weeks, and keep shipping until it does the whole job. No handoff to a delivery squad, no requirements document that goes stale in a drawer. The three systems below were built that way, and every one of them is in production.

Open-source technology radar

Nodus

A self-hosted technology radar for teams that have outgrown a spreadsheet of things we're watching. An interactive radar, a structured catalogue of technologies and assessments, role-based access with Entra ID single sign-on, and reproducible exports for stakeholder reporting. Open source, so the scouting data never leaves the organisation that produced it.

In use at TenneT and Gasunie

PythonFastAPID3PostgreSQL

Bookkeeping integration for a healthcare practice

From patient file to ledger

A small integration for a healthcare practice: patient invoices from the electronic record system and card payments from the terminal land in e-Boekhouden without anyone re-typing them. What used to be an afternoon of copying every week is now a review of the result. Small on purpose: no new bookkeeping package, just the missing link between three systems the practice already had.

An afternoon a week, returned

Pythone-Boekhouden API

Also shipped

Fast, accessible, EU-hosted websites — designed, built and operated end to end.

How we work

Four steps, and you hold more after each

AI work differs from ordinary software on one point: the model is sometimes wrong. Every step below is built around that.
  1. Pick the work

    Not “where could we use AI”, but which queue, which decision, which afternoon. We read your real cases first, and we say so when a script would do.

  2. Measure first

    Before a system exists: some twenty real cases, each with its right answer. From then on progress is a number you check, not a demo.

  3. Build

    Short cycles, running at yours from week one. When the system is unsure, a person decides, by design. The test set runs on every change.

  4. Hand over

    Repositories, infrastructure as code, test set and documentation, in your accounts. That test set lets you swap the model later without us. If we keep running it, that is your choice.

You hold

  • written plan from step 01
  • scored test set from step 02
  • running system from step 03
  • the keys from step 04

Settled at the start

The parts that are hard to change later

Four things we settle in the first week, because after delivery they are expensive to retrofit.

You own everything

Code, data and infrastructure sit in your accounts and your repositories from the first week, not in ours. No proprietary runtime, no licence to expire. Untangling this afterwards is a project of its own; at the start it is a setting.

How we work, step 4: the keys

Hosted in Europe

Where data lives decides which law it falls under, and moving it gets dearer every quarter: more data, more integrations, more people who know the address. So we pick the region before the first deploy. On ours or on yours, and that choice stays reversible.

All of our hosting is inside the EU

You talk to the builder

Nobody sits between the problem and the person fixing it. The one who scopes the work writes it, and is still the one you call later. Knowledge of your cases lives in one head you know, not spread along a chain of handovers.

Your message lands with an engineer

Open where it can be

What is not your competitive advantage, we build on open source and contribute back. That is a decision for the first line of code: opening a closed core later means untangling everything that grew around it in the meantime.

Nodus: open source at TenneT and Gasunie

Contact

Tell us what you are trying to build

A first conversation costs nothing and usually takes half an hour. We reply within one working day.

A few sentences is plenty. What exists today, and what should exist instead?