The sixth Generative AI Belgium Meetup brought the local community together around a practical question: how are teams turning rapidly changing model capabilities into products people can actually use?
The event took place on 8 May at Upgrade Academy in Ghent, with an online option for remote attendees. The relaxed format created room for technical presentations, product discussions and conversations between people working across different parts of the AI ecosystem.
Four views of applied AI
Xander Berkein presented Dealside's approach to using AI in business operations. The session focused on translating model output into a product workflow rather than treating the model as the complete product.
Lucas Belpaire discussed Bizzy's AI capabilities and its work with Google's PaLM models. The presentation showed how established data products can add generative features when the surrounding context and user need are clearly defined.
Raphaël Vorias introduced Glif.app and demonstrated image generation through composable workflows and API calls. His talk highlighted the creative possibilities that emerge when generation becomes a building block that users can connect and reuse.
Fine-tuning with purpose
Tiebe Parmentier closed the talks with a clear explanation of large-language-model fine-tuning. He made the technical ideas accessible without hiding the trade-offs. Fine-tuning can adapt behaviour to a specific domain, but it requires disciplined data preparation, evaluation and a reason to prefer it over prompting or retrieval.
Model customisation should start with a measurable failure in the current system. Without a baseline and evaluation set, improvement is only an impression.
The value of a local community
The networking session was not an afterthought. It gave attendees the chance to compare implementation details, challenge assumptions and connect the talks to their own projects. Those conversations made the evening more useful than a one-way sequence of presentations.
The meetup reflected a healthy shift in the generative AI conversation. Novelty still attracts attention, but practitioners are increasingly focused on integration, reliability and value. The most interesting work is no longer proving that a model can generate something. It is designing the data, evaluation and product context that make the generation dependable.
Ghent has a growing group of people working on exactly those questions. Events like this help turn individual experiments into a community of practice.