Conveners
AI, but we Kept the Manual: 7B
- Joost Gadellaa (SURF)
Description
This session shows how universities and NRENs are building AI they actually control and own for several use-cases we all recognise. Across chatbots, advisory systems, agentic workflows, and speech-to-text infrastructure, the talks focus on practical ways to keep data, governance, and decision-making in-house. Rather than relying on commercial tools, these examples use open standards, shared infrastructure, and human-in-the-loop design. These AI systems are transparent and fit for real usage in education and infrastructure. Let's talk about how we can keep ourselfs knowledgable and in control of these technologies.
Across Europe, municipalities face increasing pressure to deliver secure, privacy-compliant, and accessible digital services for education. These challenges are amplified by fragmented responsibilities, limited resources, and rapid technological change. By creating a centralized service and developing shared frameworks, automated support tools, and AI-powered advisory systems , we aim to...
Generative AI is rapidly entering higher education, yet most institutions rely on commercial, opaque systems that undermine autonomy, privacy, and pedagogical control. Karlstad University has developed a system for digitally sovereign AI chatbots built on open standards, open-source technologies, and institutionally governed infrastructure. Integrated via LTI and using WordPress as a modular...
Speech-to-text is rapidly becoming a core capability in teaching, accessibility and research, but many institutions struggle with trust, cost and data protection when relying on commercial AI services. This talk presents Sunet Scribe, a national, self-hosted Whisper-based speech-to-text service operated inside the Swedish NREN. Starting from concrete operational experience, we explain why...
Agentic AI systems can plan actions, call tools, analyse logs, and trigger real automation. While this enables powerful operational workflows, it also introduces serious risks: hallucinated decisions, loss of control, escalating costs, and unclear accountability. This presentation introduces Saz, a self-hosted agentic workflow engine designed to embed AI into operational environments through...
