Generative AI Workshop at UMZH Summer School

Generative AI has rapidly transitioned from an abstract research topic to a technology shaping industries across health, science, and business. To bridge the gap between technical foundations and practical applications, I led a 90-minute workshop on Generative AI at the UMZH Summer School for an audience of around 50 participants. The group was highly interdisciplinary, spanning fields from neuroscience to computer science, which made it an ideal setting for cross-disciplinary learning.

Me talking about interesting stuff

The Challenge

Teaching Generative AI to a diverse audience requires balancing technical rigor with accessibility. Participants ranged from those deeply familiar with neural architectures to those encountering AI for the first time in their domain. The workshop therefore needed to demystify the fundamentals of large language models (LLMs) while keeping the content relevant for real-world applications across research areas.

The Approach

The session began with an introduction to basic GenAI concepts—what large multimodal models are, how they process inputs, and why transformers underpin their success. Using simplified examples, I illustrated how models predict the most likely next word, explained the role of sampling and temperature, and discussed limitations such as hallucinations and bias.

The presentation then transitioned into practical techniques for prompt engineering, showcasing how role-playing and contextualization can drastically alter outputs. Participants saw live examples where the same question, framed differently, produced responses aligned with roles such as doctor, researcher, or consultant.

Finally, I introduced multi-agent approaches where multiple AI systems critique and refine each other, improving accuracy and safety—an emerging frontier in AI reliability.

Hands-On Demonstration

To ground the theory in practice, I showcased a custom-built web application (see screenshot below) that interfaces with my own MCP-powered personalized media generation server. The system allows users to create personalized podcasts and videos by configuring evidence-based inputs, conversational style, visual design, and output format.

PRECIOUS Content Generator

Participants experimented with the system by generating podcasts in real time, observing how their prompts and parameter choices influenced both style and content. This live demonstration highlighted both the creative potential and the responsibility that comes with deploying such technologies.

Key Insights

  • Accessibility matters: even participants without technical backgrounds could grasp core ideas when explained through analogies and demonstrations.
  • Interactivity is crucial: engaging participants with hands-on tools sparked curiosity and deeper understanding.
  • Cross-disciplinary dialogue: neuroscientists, computer scientists, and clinicians each raised unique perspectives on how GenAI might be integrated into their work.

Looking Ahead

The workshop underscored the importance of making Generative AI approachable across disciplines. As the technology continues to evolve, its transformative potential will hinge on whether researchers, clinicians, and engineers can work together to design safe, meaningful, and human-centered applications.

This workshop was only the beginning. By combining theoretical grounding, practical tools, and real-time interaction, we opened a shared space where diverse fields could imagine the future of AI together.