Have an AI chatbot built

Building an AI chatbot or AI assistant involves more than switching on a language model. You decide which questions it should answer, connect the sources, test the answers, integrate it with your systems, address privacy and agree on how it will be managed.

We build this as part of generative AI. You can find all our services on the services page.

Usually through retrieval-augmented generation (RAG): for each question, the chatbot first finds the relevant passages in your documents, and the language model formulates an answer based only on those passages. That is how Anthony Chat, which we built for Fokker Services Group, works.

Read how it works technically in our blog post about internal LLMs and RAG.

With a fixed set of real questions and their correct answers, put together with the people who currently answer those questions. For each response, we check whether it matches the source, cites that source and says it does not know when the answer is unavailable.

We run the test set again whenever the sources, instructions or model change.

Yes. Dex, the chatbot we built for PostNL, connects to existing systems so the knowledge base does not need to be stored and maintained in multiple places. An integration can also retrieve information, such as the status of an application, using the user’s access rights.

Automatically categorising or routing questions falls under NLP and text analysis.

You decide this during development: which language model to use, whether to run it in a cloud environment or on your own servers, whether conversations are stored and who can see them. For NFZ, we worked with Microsoft Azure, among other things. If the chatbot processes personal data, the GDPR applies and a DPIA may be needed.

Read what this involves on our DPIA for AI page.

Yes. The transparency rules of the AI Act have applied since 2 August 2026. According to the Dutch Data Protection Authority, an AI system such as a chatbot must tell you at the start of the conversation that you are talking to AI, in writing, visually or through audio. Exceptions apply.

Your legal adviser can interpret the legal requirements. We build the notification into the chatbot.

Sources change: contracts, regulations and product information. At NFZ, the most important information used to be in an annual pharmacy booklet, which could become outdated during the year. Maintenance means updating sources, rerunning the test set, reviewing logs and maintaining integrations. Your team can do this, or we can.

If you would rather have a dedicated engineer on your team, see hire an AI engineer.

When the questions rarely come up, the answer is already available in one easy-to-find place, or every question requires a bespoke response from a member of staff. In those cases, a better FAQ, search function or form will help more.

If your main focus is internal knowledge in SharePoint and Teams, take a look at Joost, our internal AI assistant.