Long-form reading fosters readings skills, comprehension, critical thinking, and empathy but also demands sustained attention, making interruptions common and disruptive. Existing supports try to mitigate the disruptive effects, yet few focus on long-form content, and personalization remains limited. With advances in AI, new opportunities for adaptive, engaging support emerge. Among different types of long-form reading, this project focuses on scientific paper reading—an especially demanding context due to its dense language, complex logic, and task-driven nature.
This project addresses the question: “How can we design interactions that leverage AI to better support readers in managing interruptions during long-form reading?” Using a user-centred, Double Diamond process, I combined a diary study, co-creation workshop, and user testing to investigate user needs and prototype solutions.
The outcome is PaperCast, an AI-powered podcast tool integrated into scientific paper reading. User tests indicate that PaperCast can increase motivation and help re-engagement after interruptions. And in order to build trust, while highlighting the importance of ensuring AI outputs remain source-traceable and carefully balanced to prevent information overload.
Beyond the design itself, the project shows how AI’s uncertainty, adaptability, and generative capacity can be treated as design materials. Insights include a framework linking interruptions and reading experience from user study, strategies for constraining AI outputs, and lessons for integrating AI into design. Future work should iterate the prototype, assess effects on reading comprehension , and explore broader contexts and modalities.

Clinical communication and information management are essential to healthcare, yet they remain challenged by fragmented documentation, fast-paced exchanges, data loss and burdensome administrative processes with Electronic Health Record (EHR) systems. Meanwhile, Artificial Intelligence (AI) shows strong potential to address these challenges by enhancing workflow efficiency and supporting real-time information documentation.

This graduation project, guided by TU Delft and Accenture IXID, investigates how AI can be integrated into healthcare communication processes. Literature review, comparative case studies of intensive care units (ICU) and nursing homes, and user research identified six shared pain points. These insights informed key design opportunities and criteria.

The final design proposes an AI-driven companion system, consisting of an online platform and a wearable voice-enabled pin. The pin enables privacy-conscious real-time voice capture and seamless authentication, while the platform supports transcription, classification, OCR-based digitization of handwritten notes, and AI-assisted report generation with traceable data sources. Together, they serve as a bridge between HCPs and EHR systems, reducing information loss and clerical burden.

Evaluation sessions with HCPs, patient families, and design experts confirmed the concept’s value and feasibility, while highlighting opportunities for refinement. This project contributes practice-based insights into designing AI systems that embed naturally into healthcare workflows, enhancing efficiency, reducing administrative strain, and supporting better communication outcomes.

This Master's Thesis examines cognitive load and usability heuristics in simulation-based training, specifically through a case study of a medical training platform developed by Laerdal Global Health, known as LIFT Scenarios. The study aims to determine the extent to which usability heuristics help reduce the cognitive load experienced by observers while using LIFT Scenarios during medical simulation training.
Following the Improving Improvement model, the research systematically addresses the behavioural and user interface factors that influence cognitive load throughout the application. Each design phase employs various methods, including user tests, interviews, heuristic evaluations, and prototyping. A total of 20 participants were involved in the research activities, excluding observations. The application of usability heuristics positively impacted the ranking of the prototypes. However, a significant portion of the cognitive load associated with LIFT Scenarios is intrinsic, stemming from task difficulty or the skills of the facilitator. This research utilizes quick testing and evaluation methods that are both feasible and adaptable.

This thesis project began with a holistic exploration of fiction reading and the emerging development of immersive reading in virtual reality (VR). The investigation focused on the concept of narrative transportation—a mental state in which readers feel absorbed into a fictional world and may lose track of time in the real one. One key challenge in the context of VR fiction reading is the lack of clarity around what enables transportation in such environments. This research takes that uncertainty as its starting point, aiming to uncover the sources of narrative immersion and explore how they can be amplified through design in VR contexts.

The early phase of the project involved a user study with eight participants who regularly read fiction. Semi-structured interviews were conducted to understand their reading contexts, the sources of transportation across different media formats, and their perceptions of VR fiction reading using a 2D demo video adapted from previous studies. To analyze the abstract qualities of transportation, the Evaluation Grid Method (EGM) was applied—an effective technique for uncovering layered emotional responses and design preferences.
Key insights from the user study revealed that readers often create or seek environments helpful to immersion, and that visual cues—such as book covers or illustrations—serve as anchors that help them re-enter the fictional world and revive mental imagery. These insights led to a refined design goal: “Amplify narrative transportation in VR fiction reading by leveraging environmental and visual cues beyond the text itself.”

Through brainstorming and storyboarding, several design concepts were proposed, focusing on “transporting readers to other locations” and “interacting with symbolic objects that reflect previously read story elements.” Multiple rounds of low-fidelity prototyping were conducted, leading to the final research prototype designed to evaluate how different visual and environmental cues affect narrative transportation. The main design variables included: (1) Google Street View vs. AI-generated panoramic images, (2) static vs. dynamically shifting imagery that conveys spatial or temporal progression, and (3) whether showing an AI-personalized book cover—based on readers’ previous responses—enhances re-entry transportation.

In the evaluation phase, four participants engaged in two sessions on different days. The first session compared the effects of Real vs. AI-generated images and Static vs. Dynamic transitions. The second session evaluated whether an AI-personalized book cover, generated from the participant’s previous reading experience, enhanced transportation upon re-entry. Results showed that AI-generated imagery generally led to higher transportation scores than real street view images, as AI visuals could better match the story’s narrative content. Similarly, dynamic imagery that progressed with the narrative yielded higher immersion than static scenes. However, the personalized book cover had limited impact on re-entry immersion, as participants struggled to relate to the AI-generated symbols and found them weakly connected to the ongoing reading experience.

The findings highlight the potential of AI-enhanced visuals in immersive VR reading. When the background visually resonates with the narrative, it can significantly support transportation. At the same time, the study reveals a potential tension between AI-generated visual cues and the reader’s desire for imaginative freedom—a core motivation for fiction reading. Future research must investigate the optimal level of visual abstraction that supports immersion without undermining creative interpretation.

While the current VR prototype was relatively static, this project points to broader affordances of VR in enabling multisensory and immersive experiences. The outcomes are forward-looking, offering insights into the evolving intersections between AI, VR, and reading. In an era where attention spans are fragmented and reading motivation among younger generations is declining, immersive reading formats like VR could offer new ways to revive narrative engagement.

Future directions include exploring how reader profiles, fiction genres, and physical reading environments may affect the effectiveness of AI-supported immersive reading. This thesis contributes early groundwork toward that vision.

This thesis presents the design and development of an Educational Toolkit aimed at building the resilience of 15–16-year-olds against AI-generated extremist disinformation. With the rise of generative AI, AI-powered tools are rapidly increasing in sophistication and accessibility. Malicious actors, mainly extremist groups, leverage these technologies to manipulate, recruit, and radicalise youth through spreading disinformation by various tactics. Examples are the generation of deepfakes, memetic warfare, and AI-enhanced grooming. Existing media literacy interventions have not yet answered to these developments and a digital media literacy tool for this specific target group is yet to be made. This thesis aims to respond to this need.
To address this gap, an educational toolkit was designed, combining Inoculation theory with an interactive, gamified experience. The contents are structured through various frameworks. The toolkit will consist of multiple lessons, where each lesson is paired with a serious game that simulates real-world disinformation tactics in a controlled and ethical environment. These lessons will focus on key manipulation techniques such as meme-based manipulation (memetic warfare), source impersonation through AI, and emotionally charged misinformation campaigns.
The thesis uses Bloom’s Taxonomy to define cognitive learning objectives and Gagné’s Nine Events of Instruction to guide lesson structure. Integrating the theory with the game, and connecting multiple lessons to form a reflective experience, is done by using Kolb’s Experiential Learning Cycle. This theoretical foundation is operationalised in the game through the Mechanics-Dynamics-Aesthetics (MDA) framework. One of the toolkit’s most notable features is a genAI sandbox, a simulated chatbot and image generation interface that allows students to experiment with AI prompt creation in a fictionalised and safeguarded setting. This component encourages hands-on learning and enables learning by doing.
Two prototypes were developed: one using Twine and one via ChatGPT’s custom GPT function. A guest lesson with a live classroom gave insights on the target group and shifted the project focus to educational environments instead of standalone serious games. During the lesson, students created AI-generated memes containing disinformation. This activity was followed by reflection on narrative techniques and emotional impact. A general survey and classroom observation further supported the need for not only gamified approaches, but also teacher-led theory and reflection.
While the educational toolkit lays a strong foundation, further steps are needed to bring it into practice. These include further developing of the lesson theory and corresponding serious games and validating the toolkit in high-schools. Completing these steps will turn the educational toolkit from a high-level research-based solution into a digital media literacy solution for strengthening resilience among the target group.
This thesis contributes to the field of design for digital resilience by offering a proposed, research-based educational toolkit that acknowledges the current and evolving threat landscape as enabled by generative AI. It also provides a replicable framework for integrating serious games into (digital) media literacy education.

With the European population rapidly aging, addressing the vulnerabilities associated with physical and cognitive decline is essential. Promoting an active lifestyle among older individuals has proven effective in preserving health and independence. Growing evidence suggests that combining cognitive and physical exercises can significantly enhance both cognitive function and physical health, especially among sedentary elderly populations who are at increased risk due to physical inactivity.

This thesis, conducted in collaboration with the Portuguese company Neuroinova, presents the development and evaluation of a digital intervention aimed at integrating physical exercises into their platform for cognitive training COGWEB in individuals experiencing cognitive decline or impairments. The integration process started with ACTIVAS, a program developed in collaboration with researchers from the University of Aveiro, who provided guidelines for an effective physical exercise regimen based on personalization and progression. However, motivating users to adhere to the program, particularly in a remote setting, remained a challenge.

Understanding the needs and barriers of both patients and professionals was crucial to shaping a program that positively impacts user motivation.
Utilizing a user-centered design approach, including generative sessions, interviews with healthcare professionals, and literature reviews, two primary user groups were identified: sedentary individuals with minimal physical activity and busy active individuals with irregular physical exercise patterns. Despite individual differences, the insights revealed that successful integration of physical exercises into COGWEB requires clear, structured guidance from a trusted professional, such as a physiotherapist, alongside engaging and enjoyable exercise activities. Personalization of the program emerged as a key factor in addressing these needs.

The design process focused on ensuring effective guidance and engagement by iteratively testing various prototypes of exercise sessions, with special attention to interaction qualities such as being engaged, guided, proud, trusting, and in control.
The final concept involved a session where cognitive tasks in the form of games—similar to the original COGWEB format—were paired with physical movements and exercises, allowing users to train both cognitive functions and their body simultaneously. Emphasis was placed on providing clear instructions and professional guidance through the presence of a physiotherapist.

Final test results showed promise in motivating sedentary individuals to participate in these sessions, achieving the design goal of ensuring guidance and engagement. While the design objective of this thesis was met within the study's scope, further work is necessary to implement and evaluate the program's long-term effectiveness in real-world settings. The findings suggest that this approach has the potential to significantly increase physical activity levels and improve adherence among elderly COGWEB users, contributing to their overall well-being.