provisional-pilot-diy-bot-workshop
Before the session
send a pre-workshop email with Ollama installation instructions and a link to pull one small model (gemma3 or llama3.2 – something that runs on modest hardware). People who arrive with it installed move faster; people who don’t can install during the first section while I am talking.
The room
Laptops out, projector showing mine. run everything live so participants can follow along and see what I am doing in real time.
Opening (15 mins)
talk through the critical framing – not a lecture, more like a provocation. Why local, why now, what’s at stake. show my own setup briefly: the supervisor bot, the study companion. The point isn’t to impress but to show that this is buildable and that I hace done it myself. ask the room: what shape do you imagine an AI tool taking in your own practice?
Dreaming (15 mins)
Participants fill in a worksheet – designed to feel more like a design brief than a form. Four questions: what would you call it, how does it speak to you, when would you use it, what’s the one thing it should always or never do. This is the ideation moment. I will circulate, ask questions, encourage people to think beyond chat.
Building (25 mins)
walk through the scaffold live – a simple Python script or a pre-built interface where they drop their four decisions in and the bot appears. do it first on the projector with a demo bot, then they do it on their own machines. I will circulate. People who finish early can start experimenting with adjustments – changing the system prompt, trying different instructions.
Exploring alternative forms (15 mins)
show two or three quick demos of non-chat implementations: a voice-activated version, something that reads from a folder of notes, a bot triggered by a button rather than a typed prompt. don’t build these in the session – just show them running. The point is to open up the possibility space beyond what they’ve just built.
Reflection and close (15 mins)
Group conversation: what surprised you, what would you change, what’s the bigger version of this you’d want to build. point them to resources for continuing – the repo they’ve cloned, where to find models, how to keep modifying the system prompt. Brief mention of the political stakes again: ‘this is yours, it runs on your machine, nobody else has access to it’.
The revisions below show how this note has changed over time.