AI WHERE IT HELPS

A custom AI chatbot for your own documents

A chatbot that answers from your own manuals, rules and website. Not from the whole internet. Connected to the AI model that suits you.

When your own chatbot makes sense

  • The same questions keep coming back to your team or your office.
  • The answer is in a document somewhere, but nobody remembers where.
  • You want to use AI, but your data cannot simply go to a service outside Europe.

How it works

Your own sources

Documents and web pages are read in and made searchable. The chatbot first looks up the passages that match the question, then answers. This is called RAG.

The model that fits

OpenAI, Anthropic or a model on your own servers. Want to keep your data in Europe? Then I choose a model in a European cloud region or on your own servers. I choose what fits the question, the budget and where your data is allowed to be.

Up to date

New documents and changed pages are read in again, so the chatbot keeps up.

In practice: a platform for clubs and associations

A software supplier for clubs and associations wanted to know whether every club could get its own AI chatbot. I built a proof of concept.

Brief

Every club has its own regulations, its own rules and its own website. Members keep asking the same questions. The chatbot had to answer per club, from that club's sources.

Approach

  • A platform where every club or organisation sets up its own chatbot.
  • A pipeline that pulls in documents and websites, with a crawler for the club's site.
  • Integrations with several AI providers: OpenAI, Anthropic and local models.
  • A database spread across several servers (sharded Postgres), so the platform grows with the number of clubs.

Result

A working demo at several clubs. Each chatbot first searches its own club's sources and answers from them. That is called RAG: the chatbot bases its answer on what the club itself has written down.

Frequently asked questions

What is RAG?

Retrieval-augmented generation. The chatbot first searches your sources for the passages that match the question and passes them to the language model. That way it answers from your information, not from its general knowledge.

Does my data stay in Europe?

It can. With a model on your own servers, or in a European region of a cloud provider, your data stays in the EU. If you use a service outside Europe, you and I go through beforehand which data goes there.

Can the chatbot also do things, not just answer?

Yes. Then it is called an AI agent: it can book an appointment or create a ticket, for example. I only build that where it really saves time.

Tim Geerlings

Your own chatbot?

Tell me which questions it needs to answer and from which sources. You will hear from me.

opens in a new tab