Eva Verhelst and Mathieu De Coster recently travelled to Mzumbe University in Morogoro, Tanzania, to install a new GPU workstation and deliver a two-day workshop on running generative AI models locally for educational applications.

The visit marks a small but meaningful anniversary: the server room that now houses the workstation was inaugurated almost exactly ten years ago by the Belgian ambassador at the time, Paul Cartier. A decade later, the same room is hosting a new generation of hardware, this time dedicated to generative AI research and teaching in Tanzania.

The workshop coincided with the inauguration of CAIRI, Mzumbe’s newly launched Centre for AI Research and Innovation (CAIRI), where AIRO professor Tony Belpaeme was given the honor to cut the ribbon together with the Mzumbe University’s Deputy Vice Chancellor for Academic, Research, and Consultancy. It marks the first official collaboration between CAIRI and AIRO. The partnership sits within the broader framework of two VLIR-UOS funded projects: “AI4STEM: AI-driven Inclusive STEM Learning for Tanzania”, and “The Makings of a Global Citizen: Leveraging Generative AI for English Mastery in Tanzanian Primary Schools”.

The workshop itself focused on building intelligent tutoring systems that can run entirely on local infrastructure. This is a deliberate choice, given the bandwidth, cost, and data-sovereignty realities of deploying AI in Tanzanian classrooms. The workshop included hands-on exercises around LLMs, visual question answering, Stable Diffusion image generation, and a speech-to-text / text-to-speech pipeline, all served from the new GPU machine and integrated into small client applications. The workshop materials are available on GitHub.

This was Ghent University’s second workshop on generative AI at Mzumbe University. In August 2025, Eva and AIRO alumnus Ruben Janssens facilitated an earlier session introducing generative AI using cloud-based models. Moving from cloud APIs to locally hosted models on a dedicated GPU workstation is a natural next step. It puts CAIRI in a strong position to develop AI tools tailored to local languages and educational needs.

We at AIRO are looking forward to seeing what the CAIRI team builds next and the two research groups will continue their international collaboration.