In his Bachelor’s thesis, Robert Halbach explores application potential and limitations of implementing AI components into digital services. As a practical study, the work revolves around the digital archive of KISD – a decade-old platform containing student works dating back to the pre-2000s.
Since the launch of OpenAI’s ChatGPT in late 2022, public attention toward the fields of artificial intelligence and machine learning has skyrocketed. This attention is accompanied by a large-scale “AI race” that has, since its beginning, led to rapid advancements and a myriad of new models focused on a plethora of different purposes. While large state-of-the-art multimodal models such as Claude, GPT-4o, or Gemini form the centerpieces of public communication, research revolving around the implementation of AI models has incentivized the development of lightweight yet performant model families.
Optimized for use in resource-constrained environments, these offer opportunities to integrate into patterns like web pipelines, where lightweight and efficient solutions are crucial. Exploring potential applications within a real-world scenario, “.aaaarchive” aims to solve problems that have built up over the years within the KISD archive. This exploration resulted in a new iteration of the archive platform that featured AI functionality on multiple occasions—such as automated tag generation, cover color extraction, and a vector-based recommender system.
However, while these solutions highlight how AI can be a great tool to solve some problems, the project also showcases its limitations. Like any tool, AI components excel in specific use cases, making the decision to include them a matter of selecting “the right tool for the job.” A key lesson from this project was that the design approach to the features themselves is equally, if not more, important than the tools used to support them.
Supervisors:
Prof. Dr. Lasse Scherffig, Interaction Design
Mauro Rego (Extern)
