Skip to content

AI agency in Groningen

We build and manage applications based on language models, such as the LLM-based sentiment analysis in the social media monitor for the Respect Foundation.

For the Respect Foundation, our team used an LLM to analyse the sentiment in coverage of the Week van Respect

Through our Data For Good initiative, we set up a social media monitoring tool and carried out an impact analysis. This gave the Respect Foundation insight into the sentiment and actual reach of its annual campaign.

Read the case study

How much text do we receive every day that no one has time to read in full?

Responses, requests, notifications and messages about your organisation: in local government, education and healthcare in Groningen and the North, the volume is growing faster than teams can keep up. A language model reads alongside your staff, organises and summarises; your staff make the decisions.

Book a meeting

Language models that read, assistants that answer and models that classify, built and managed

Discuss your AI application

Reading social media posts

Around Week van Respect, we set up a social media monitor for Respect Foundation; an LLM determined the tone of each post.

An assistant using your own sources

For NFZ, we built a virtual assistant using information from contracts with health insurers, which pharmacists consult at the counter.

Classifying and matching

For Olympia, we developed a classification model to match candidates with suitable job openings more efficiently and accurately.

Putting models to work

For LUMC, we set up the infrastructure that enables its own ML models to work in practice; the results appear in a clear, easy-to-use application.

Your organisation needs to be able to oversee what we build remotely

  1. One stream of text or one task

    Together, we identify where most of the reading or searching happens and who in your organisation will take ownership of the application.

  2. Test with users

    The employees currently doing the work assess the results against a fixed set of examples from their own work.

  3. Management you can see

    Logging, monitoring and a standing agreement on making adjustments. We handle this, or your own team takes over.

Can an Amsterdam-based AI consultancy work for us in the North?

Find out in a conversation

A good fit if

You handle a lot of text, or colleagues keep searching through the same regulations and contracts. Everything is available digitally, and a designated colleague takes responsibility for putting it to use.

Not a good fit if

The task comes up only a few times a year, or the sources contradict each other. In that case, tidying up the sources or using a smarter fixed rule will be more useful. We’ll tell you that upfront.

What determines the scope

The number of types of text or tasks, the systems we connect, who can access which data, the environment it runs in and whether your team takes over its management.

Are we tied to one language model?

Preferably not. Code, prompts and test questions stay in your own environment, and we agree on the model and hosting together. That makes it possible to use a different model later.

Organise text, answer questions or take action: where do you start?

Read and organise text

Choose this if you receive a large number of messages or requests every day. Our advice for a first application: let the model label and summarise them, while a member of staff makes decisions and corrects errors.

An assistant based on your own sources

Choose this if colleagues keep searching the same policies, contracts or manuals. Every answer shows where it came from, so anyone can check it.

An agent that takes action

Choose this if work repeatedly follows the same path through a few systems. Define in advance what the agent can do without human involvement.

From Amsterdam, we’ve been building AI applications well beyond the Randstad since 2016

Data Science Lab is a Databricks and Microsoft partner and works from Westerdoksdijk. We built applications based on language models for the Respect Foundation and NFZ. We share what we learn along the way at our Techdays.

Book an introductory meeting

In our first conversation, we decide which text or task should get a language model first

Stef Stoelwinder

Stef Stoelwinder

Business Development

Stef is your first point of contact. In our first conversation, online or at your premises, we discuss which text or task would benefit from an AI application first.

Which stream of text or task would you like AI to prepare first?

Tell us what text you receive and which systems it ends up in. We’ll reply by email or suggest an initial conversation.

Enter a valid email address.

Let us know how we can help you.

Frequently asked questions

We build applications that read and organise text, assistants that find answers in your own sources, and agents that carry out steps in a process. We test them with your staff and continue to maintain them afterwards.

If you’re looking for an AI company or AI agency in Groningen, this is the work we do. You’ll find the full overview under AI applications.

No. We work from Westerdoksdijk in Amsterdam. We work online with organisations in Groningen, Friesland and Drenthe, and visit when needed, for example to test with users.

None of the case studies here took place in Groningen; they show what we build. For advice on AI in networks, water and buildings in the north of the Netherlands, see AI consultant Groningen.

By comparing it with your own staff. Staff label a set of messages by hand, the model assesses the same set, and we examine where they differ. We run that set again after every change to the prompt or model.

For the Respect Foundation, an LLM assessed the tone of messages about the campaign; the case study identifies LLM and Python as the technologies used.

Yes. For each question, the assistant first finds the relevant passages and only then writes an answer, with a reference to the document. That lets you check what the answer is based on.

For NFZ, we built a virtual assistant using information from contracts with health insurers, supplemented with practical information for pharmacists. For more on language models, see generative AI.

Only when it’s clear which actions are permitted and every action can be traced. An agent usually starts by proposing actions, such as preparing a record, which a staff member then approves.

If your main aim is to transfer fields from forms or PDFs, see AI document processing. For more on agents, see agentic AI.

We decide that with your IT team and privacy officer: in your own Azure environment or somewhere your organisation already uses. Microsoft has an Azure region physically located in the Netherlands.

If the application processes residents’, students’ or patients’ personal data, a DPIA is often needed. For what we provide to support that, see DPIA for AI.

We agree on that before starting: who owns it, who gets alerted to unusual results, and a set process for releasing a new version. Sources and models change, so ongoing maintenance is needed, even remotely.

For how we monitor models, see MLOps. We also provide ongoing maintenance through management and continuous improvement.

This is about applications used in day-to-day work: language models, assistants and agents, plus their maintenance. Our data science agency in Groningen builds models that predict or classify.

If you mainly need reliable reporting across locations and departments, see Power BI specialist Groningen.