Knowledge and prototypes
Build your own chatbot
The same questions come up again and again: what does it cost, when can you deliver, how do I do this. The answers already exist in your own material — they are just hard to find, and someone has to write them out every time.
On this full-day course you build a chatbot on your own content. We start with the questions you actually get, gather the material that answers them, and have the first version answering before lunch.
The afternoon is about what separates a usable bot from an embarrassing one: a tone that sounds like you, clear limits on what it will speak to, and a decent answer when it does not know. We test with the hard questions and build a fixed list you can rerun after every change.
At the end of the day we cover running it and being responsible for it: where the bot should live, what it costs, what you are liable for when it answers in your name, and what GDPR requires. You go home with a working prototype and a plan for putting it into service.
Programme
1. What should the chatbot answer?
We start with the questions you get again and again: from customers, colleagues or citizens. Which of them can a bot answer, which should it leave alone, and what does a good answer look like.
2. How a chatbot on your own content works
Why a language model needs your material handed to it in order to answer from it. We draw the path from document to answer — and why the bot must cite its sources.
3. Collect and prepare the content (hands-on)
We start from your own material: FAQs, guides, product sheets, terms. What works as it is, what needs rewriting, and what to keep out. Poor content gives poor answers — this is where quality is decided.
4. Build the first version (hands-on)
From content to a bot that answers. We set it up with Claude, give it your material and ask the first real questions — the ones you got last week.
5. Give it a role and a limit (hands-on)
Tone, length and phrasing so the answers sound like you. And the crucial part: what the bot should say when it does not know — instead of guessing. We write the rules and test them.
6. Test with the hard questions (hands-on)
We try to make the bot fail: off-topic questions, attempts to make it promise something, and questions where the material says two different things. A fixed list of test questions you can rerun after every change.
7. Put it into service (hands-on)
Where the bot should live: internally for colleagues, on the website, or both. What it costs to run, how to keep an eye on usage, and who takes over when it cannot answer.
8. Responsibility, data — and what you do tomorrow
What you are liable for when a bot answers in your name; personal data in questions and logs; GDPR and the duty to inform. You go home with a working prototype and a plan for putting it into service.
Upcoming dates
No dates are scheduled for this course at the moment. Register your interest below, and I will contact you as soon as a date is set.