Professional learning for member companies
Short videos about tools, methods and experience. Descriptions are public; the videos themselves require access through company membership.
Watch the videos before the discussion meeting.
Sprint 3
16 Nov 2026 – 29 Nov 2026
Publication deadline: 23 Nov 2026 · Discussion: 27 Nov 2026
Artificial intelligence · Sprint 3
My local AI architect: three models draw a flat
Three local language models receive the same task: draw a one-bedroom flat of approximately 60 m², using ArchLang to draw and validate the layout. The film reviews responses from gpt-oss:20b, Gemma 4 and Qwen3 Coder, showing how they handle the floor plan, furniture, doors and requested changes. The setup runs Open WebUI and Ollama on a TensorBook, with access from a laptop and mobile phone through Tailscale. The presentation explains the system prompt, the example supplied to the models and the stopping rules for tool use. The demonstration reveals both possibilities and concrete errors – and why a result still needs to be checked against the brief.
View videoAI · Sprint 3
More than a model: language model, agent and harness
For anyone who uses AI tools in a chat window or a terminal and wants to understand what happens behind the answer. You learn the difference between a language model, an agent and a harness: the model only produces text, the harness is the program around it that gives it memory, instructions and tools, and together they make an agent. In a demo in DeepSeek Harness, which is open source, we follow a message through the harness step by step. A simple “hi” becomes 18,000 tokens before the model answers, and after one small web search the conversation has used 59,000.
View video