Build your own chatbot

What it reads, and who decides

Youth AI Lab (Erasmus+): six hands-on workshops and six guides for youth workers, to see what happens behind a chatbot before using one with young people.

All the workshops answer the same question: what enters the model's context, when, and who decides? The context is everything the model reads before it answers: the question, its instructions, and any document it was given.

Workshops 1 to 5 each add one piece to the chatbot of the pair, and the same test questions are asked again after each piece, so that everyone sees what changed. After workshops 1 and 2, the chatbot can already be used: workshop 1 gives it its instructions, workshop 2 its knowledge. Workshop 5 closes the series; workshop 6 goes further, with a small model that sorts the messages before the chatbot. Partners can then run the WikiYouthBot activity with young people (12–15) and know what is behind the two boxes, Instructions and Knowledge, of the tools they use.

The six workshops

1Instructions, and how to test a chatbot

What does the chatbot read before it answers?

A test set of questions and a scoring rule; five chatbot personalities; your own system prompt; your chatbot in Open WebUI.

3Let it search, give it know-how

What if the model searched by itself?

The model searches the files itself; a document of your territory enters the knowledge, checked by a person; your own skill; your chatbot in Open WebUI.

4Plug in tools

What happens when the chatbot acts outside?

A tool that gives the fire danger level of the day (MCP), with its official source, and what a tool can do wrong; your chatbot in Open WebUI, with its tool.

5Organise what it knows

How do you organise knowledge so that a person and a model can explore it?

A wiki that the model keeps and a person checks; a knowledge graph; the chatbot of your territory, packed to rebuild elsewhere.

6Teach a small model to sort messages

How can a small model that you teach help a big one?

A small multilingual model, fine-tuned on your own examples to sort the messages before the chatbot, then compared with a general chat model.

Home of the series: the kit

Everything the first five workshops use, in one place: setup, commands, common problems, scoring rules, how one workshop leads to the next.

Each repository has its page (the workshop, step by step) and a one-page sheet, written to be translated. In the slides: arrow keys to move, f for full screen, p for the presenter view.

Six guides, to go further

Open WebUI

One chatbot per topic, activity or project, shared with a group during a session; templates of instructions with rules to protect young people.

OpenCode

A chatbot of your own with another open-source tool, and five ways to pay for the model.

OpenRouter

The other choice to an Anthropic key: one key, many models, no Claude model by default, and what zero data retention and hosting in Europe take.

Dev containers

The workshops and the guides on your own computer, without a Codespace.

Hermes

An assistant that remembers, to help youth workers prepare their sessions. Never for use with young people.

Where to start

You want to…Start with
Run the series with a group of youth workersThe facilitator sheet of workshop 1, then the setup of the kit
Follow a workshop on your ownThe repository of workshop 1: its page opens in the Codespace
Work without a CodespaceThe dev containers guide
Give a group a chatbot during a sessionThe Open WebUI guide, or the local AI guide to keep the data in your organisation
Build a chatbot with young peopleThe WikiYouthBot activity
Understand what a language model doesHow a language model works: tokens, the context window and temperature, with pictures
Understand a word of the workshops and guidesThe glossary: each word in plain language, with links to where it is explained

What a session needs

With young people: WikiYouthBot

In the WikiYouthBot activity, a group of young people picks a local issue, gathers the real sources, and turns a generative AI into its own expert. The series shows what is behind it.

The activity sheet

The ideation sheet of the activity (PDF, 26 pages) and its printable materials.

A real week in La Rochelle

The chatbots and robot prototypes built by young people in July 2026, among them Feurisson, the wildfire-prevention chatbot that inspired the running example of the series.

Step by step, in French

The WikiYouthBot page of the wiki of the L.A.B. (Aix-en-Provence).

Reuse this page

Link to it, or copy this short text into a report or a website:

Build your own chatbot: what it reads, and who decides is a series of six hands-on workshops of 2 hours for youth workers, made in the Erasmus+ project Youth AI Lab. Working in pairs in the browser, participants see what a chatbot reads before it answers, and who decides what it reads, by adding one piece at a time to their own chatbot: instructions, knowledge, search, tools and a wiki, then fine-tune a small model of their own. Six guides go further, for example a chatbot shared with a group, or one that runs on a computer of the organisation. All the material is free to reuse, under open licences: youth-ai-lab.github.io/yal-workshop/resources.html