This master’s thesis sits at the intersection of conversation as data and Natural Language Processing (NLP) as a method — asking what hidden insights might emerge when we treat everyday dialogue not just as noise or narrative, but as a space where values are shaped, shared, and sometimes lost. In collaborative, high-stakes settings, decisions are rarely made on logic alone; they are entangled with the values of those at the table — often unspoken, often misaligned.

The research introduces the Value Expression Gap — the disconnect between what people claim to value and what actually surfaces in how they speak. This gap became both a conceptual lens and a design target, shaping the development of a low-fidelity NLP prototype that detects value cues in conversation using sentence embeddings and semantic similarity scoring.

Through a Research through Design (RtD) process — spanning prototyping, real-world deployment, workshop observation, and iterative refinement — the project explored how values appear implicitly in tone, metaphor, emotional framing, or silence. Each stage contributed to improving how values could be surfaced computationally without flattening human nuance.

As the method was tested with organizational leaders and decision-makers, new forms of relevance emerged. Making values visible not only helped reflect on alignment and culture — it opened space for more strategic dialogue, tension awareness, and ethical negotiation. The tool was seen not as a truth-teller, but a reflective companion — prompting deeper questions and exposing patterns otherwise missed.

Rather than offering final answers, this thesis proposes a shift: toward tools that reveal, not resolve; that prompt reflection, not prediction. In doing so, it opens up a new role for AI in design and decision-making — one grounded in transparency, human judgment, and the evolving language of values.

As consulting firms face increasing pressure to deliver proposals faster and more efficiently, the RFP workflow has become a critical yet strained process, especially within high-volume environments like Accenture Song. GenAI, and more specifically LLMs, offer new possibilities to streamline repetitive tasks, improve consistency, and reduce workload. Yet, there remains little empirical insight into how these technologies can be meaningfully embedded in real-world consulting workflows without disrupting human collaboration.

This thesis explores how GenAI might support greater efficiency and employee satisfaction within the RFP workflow of Accenture Song’s D&DP team. Combining literature research with qualitative methods, including interviews, process shadowing, and a collaborative mapping session, the study identifies recurring pain points in areas such as proposal development, communication, and feedback loops. These insights are synthesised into key opportunity areas where GenAI could augment, rather than replace, existing work.

Using the Double Diamond framework, a design process was applied to translate findings into actionable interventions. Multiple concept directions were generated, evaluated, and refined through co-creation with the D&DP team. One hybrid solution was developed into a high-fidelity prototype that supports consultants in structuring proposals, maintaining consistency, and reducing rework.

The final solution is positioned across product, process, and strategic layers, offering a practical path for implementation. It shows that GenAI can be a powerful enabler, if introduced with care. Rather than automating away the human element, this project demonstrates how thoughtful design can turn emerging technology into meaningful support for the people doing the work.

This thesis explores the design and development of an integrated milk foaming solution for Quooker’s upcoming Coffee Tap — a compact, under-counter espresso system designed to minimize kitchen clutter while delivering a premium coffee experience. While the Coffee Tap enables high-quality espresso through a capsulebased system, it lacks a milk foamer — despite the fact that nearly half of consumers regularly drink milk-based coffee. Introducing a separate device would contradict Quooker’s clean-counter philosophy, making the integration of a foamer both a strategic necessity and a design opportunity. The core challenge was to create a milk foamer that seamlessly integrates with Quooker’s hardware and interface, remains compact and easy to clean, and appeals to a broad but clearly segmented user base. To address this, the project followed a structured user-centered process based on an adapted Triple Diamond model. The methodology combined quantitative and qualitative research: segmentation surveys, emotional self-tracking, in-depth interviews, and product testing. Five user segments were defined, with special focus on Automation Seekers — users who prioritize speed, simplicity, and minimal maintenance. These insights guided the selection and development of a Venturi-based solution, using steam from the Coffee Tap’s under-counter Coffee Box to draw milk through a narrowed channel, mix it with air, and create highquality foam. The device is composed of two detachable, dishwasher-safe components, and is controlled entirely via the Coffee Tap interface, maintaining Quooker’s seamless user experience. The technical development phase involved over fifteen iterations of the Venturi mechanism, each aimed at improving foam quality, air intake, turbulence, and user handling. The final lab-tested prototype produced stable, userapproved microfoam that met both technical and user criteria. A Technology Readiness Level (TRL) assessment places the solution at TRL 5, validated in a relevant environment, indicating clear potential for future integration and upscaling. Although further development is needed to reach market readiness, including manufacturing optimization, certification, and full software integration, this thesis presents a validated and strategically aligned solution that fits within Quooker’s ecosystem and meets the needs of its future coffee users.

This thesis examines the integration of generative artificial intelligence (AI) into industrial design practice, using Bugaboo's new product development as a case study. It explores the opportunities for generative AI to augment new product development, how these opportunities can be scaled, and the insights gained from its practical application in real-world design processes. Grounded in theories such as bounded rationality and expandable rationality, the work introduces the Frame–Propose–Evaluate (FPE) simplified model of design, structuring design into iterative cycles of framing, proposing, and evaluating, each supported by uncertainty-driven actions: information, representation, and reflective action.

Emerging evidence suggests that generative AI can accelerate product development by expanding the problem–solution space and compressing iteration cycles. However, it also raises concerns about increasing design fixation, limiting originality, and blurring human–AI collaboration. Its application in live, design-led organizations remains largely unexplored, offering new opportunities to study integration into uncertainty-driven workflows.

Through an Action Design Research (ADR) approach, three bespoke generative AI tools were co-developed and embedded into live design projects: InsightGPT (supporting information action during framing), CreAIte (enhancing representation action during proposing), and RulesGPT (facilitating reflective action during evaluation).

The evaluation showed that integrating AI into the slower, uncertainty-driven phases around design improved the speed and richness of iteration cycles, acting as provocation engines without disrupting the intuitive rhythm of design. In the research sessions, designers spent less time on manual research and visualization and more on framing questions, refining prompts, interpreting outputs, and aligning stakeholders. Generative AI increased design agility, stimulated divergence, and accelerated iteration.

However, important limitations emerged, including decision fatigue from an overwhelming number of AI-generated options and a critical dependence on prompt design to ensure useful outputs. Human oversight remained indispensable for interpreting AI contributions and safeguarding the quality of design decisions. New literacies, particularly in prompt crafting and AI output validation, surfaced as essential competencies for effective use.
Importantly, while AI expanded the breadth of exploration, it did not resolve the core uncertainties inherent to design; instead, it amplified designers’ ability to act amidst ambiguity. This reinforces the view that uncertainty is not a flaw to be eliminated but a creative resource to be navigated.

From these insights, the thesis formulates ten guiding principles and proposes a strategic roadmap for AI adoption in design-led organizations. Ultimately, it advocates for a mode of critical augmentation, where AI reshapes the rhythm of design while preserving its creative, empathetic, and judgment-driven roots.

This thesis explores the enhancement of active reading experiences through a human-machine co-creation process, focusing on visualized narratives. The project addresses the limitations of traditional reading materials that often neglect the individuality of readers in visualizing narratives. To bridge this gap, the project adopts a modified ZEN design approach, emphasizing user rituals and the integration of AI tools for visualizing mental models. By integrating artificial intelligence (AI) and generative technologies, the project combines human creativity with machine efficiency, aiming to create personalized digital reading experiences that cater to individual cognitive abilities and preferences.
The project begins with a comprehensive literature review that establishes the theoretical foundation for understanding narrative books and active reading. Realizing each one's active reading process and mental images regarding the same contents are different, the core of this project is to create a personalized active reading journey. Then, design vision is proposed to identify design goals and value delivery. To understand the actual user ritual, qualitative user research is conducted to study the reading activities and mental models of active readers. Then, through competitive products research, mental image visualization is found to be the active reading function currently missing in the reading products on the market. Thanks to the emerging AIGC technology, it becomes possible to generate desired visual images for people with different mental imagery abilities with high efficiency and quality. By strategically synthesizing the user requirements with image generation workflow, design round 1 focuses on content personalization. 2 solution flow was designed and tested by 10 participants with different mental imagery abilities (results from VVIQ test). After iterating the solution flow of mental image visualization function, design round 2 is conducted to realize toolkit personalization. Scenebites as an integrated active reading platform is introduced, empowering diverse user rituals of active processes. The value proposition of design vision and key user needs are validated through focus group research.
The final design output of this project is a human-machine collaborative system flow for personalized digital reading experience on knowledge construction, together with business model propositions for future libraries. The results demonstrate that personalized visualizations significantly enhance active reading experiences by reducing cognitive load and increasing engagement. And the assistance of visual options and editable mood boards in the user journey can positively match the requirements from users with diverse visual imagery abilities.
The project also suggests improvements in AI models to enhance precision and control in generating visuals. It also recommends strategies for libraries and educational institutions to integrate personalized reading services, thereby transforming the digital reading landscape into a more interactive and user-centered journey.

This project delves into the intricate challenges and potential opportunities within the Dutch hydrogen ecosystem as the country strives to meet its ambitious goal of achieving net-zero carbon emissions by 2050. The research highlights several significant barriers to the widespread adoption of renewable hydrogen, such as high production costs, financial risks, and the lack of sufficient infrastructure. These challenges have created a “chicken and egg” scenario in the market, where both hydrogen suppliers and off-takers are hesitant to commit without guarantees from the other side, stalling progress in the development of the hydrogen economy.
To address these issues, the study proposes the development of Local Hydrogen Networks (LHNs). These networks are envisioned as clusters of industries within specific regions that are connected to hydrogen suppliers, who use renewable energy sources like wind and solar to power electrolysers. By localising hydrogen production and distribution, these networks can tailor infrastructure to meet regional needs, reduce costs, and stimulate demand. LHNs also offer the added benefit of being scalable, allowing for gradual implementation that aligns with the existing infrastructure and market conditions.
A critical component of this project is the development of HySynth, an interactive online tool designed to facilitate the adoption of green hydrogen. HySynth provides stakeholders with clear, data-driven visualisations of hydrogen infrastructure, including the locations of potential off-takers and current projects. By making complex data more accessible and understandable, HySynth aims to foster greater collaboration among stakeholders, helping to align interests and reduce the risks associated with investing in hydrogen projects.
The project highlights the importance of regional collaboration and strategic partnerships as essential elements in overcoming the current inertia in the hydrogen market. By focusing on localised solutions and enhancing stakeholder engagement through data visualisation, the research contributes to the development of a more robust and integrated hydrogen infrastructure.

While robots and automation in general have been replacing humans in repetitive and routine labour for many years, the role of AI is considered to develop from mere tools to integral teammates. However, recent developments in artificial intelligence (AI) are redefining the role of AI from being mere tools to becoming integral teammates within Human-Autonomy Teams (HATs). The emergence of advanced large language models (LLMs) has extended the capabilities of AI, enabling machines to mimic cognitive human functions and thus marking a new era of collaboration between humans and AI. This transition has attracted many companies across diverse industries, such as medicine, manufacturing, spaceflight, and construction, where the development and integration of HATs are actively explored.

As the landscape of human-AI collaboration continues to evolve, there is a critical need to delve deeper into understanding how these teams will function. While research in the field of Human-Autonomy Teams has been expanding rapidly, most studies have been limited to controlled laboratory settings. Therefore, it is imperative to conduct more field research to assess the real-world effectiveness of HATs, particularly in workplace contexts.

This research focuses on exploring the potential of AI-powered systems to support team effectiveness, with a specific implementation context within ThoughtWorks. It aims to identify and address the challenges of human collaboration in interdisciplinary product development teams and examines how AI can be leveraged to enhance team effectiveness. By proposing speculative AI-driven solutions to these challenges, the study highlights not only the transformative potential of technology but also the symbiotic relationship between humans and machines in the context of teamwork.

The outcome of this research offers a futuristic vision of Human-AI teamwork scenarios and provides a framework for examining AI's role within these scenarios to augment team effectiveness. The study combines a comprehensive analysis with in-depth investigation, contributing to a holistic understanding of the roles, capabilities, and interactions of AI within teams, as well as the implications of these advancements for the future of work.

Academically, the application of participatory practices that involve the people who will use the products or services has become widely recognized and essential. However, many organizations struggle with implementing this concept within their organizational context, particularly when it comes to participatory decision-making. It is common for decisions to be made by a few individuals with authority, significantly impacting others who feel limited in their ability to influence the outcome. Consequently, there is a pressing need to develop innovative solutions to address these challenges.
This project, conducted in collaboration with CGI Netherlands GTO, aims to bridge the gap between the theoretical emphasis on participatory value and its practical application in organizations. Two research questions are formulated to guide the investigation: “How can participatory decision-making be integrated into GTO?” and “How can hierarchical structures and power imbalances in decision-making be overcome to embrace participatory decision-making?”
To address these questions, a proposed reflexivity exercise, known as Power-full Reflexivity, is introduced. The goal of this exercise is to assist individuals in positions of authority in recognizing the power dynamics inherent in the decision-making process. The ultimate objective is to facilitate the meaningful involvement of individuals who are affected by the decisions and have experienced the issues, granting them greater access and agency within the decision-making process. It is believed that the use of Power-full Reflexivity by decision-makers will enhance the participatory value in their decision-making practices, benefiting individuals through increased agency and fostering a collaborative and participatory environment within the organization.

Capabilities and limitation of topic modeling were explored through a literature review. A design opportunity was recognized in literature review for students as novice researchers. Interviews were held with master students to gain insight in their literature review process. Students reported issues with the process of related document acquisition and relevant information extraction. Document exploration was chosen as the most promising activity concerning both processes. PaperScout, a prototype for a tool which improves document exploration, has been proposed and tested with students. Recommendations have been made to further the development of PaperScout and to make other contributions to the larger process.

This research aims to create a minor taxonomy of the domain of Ambient Intelligent Environments (AmIE), sometimes simply called Ambient Intelligence (AmI). These refer to ‘smart-,’ ‘intelligent-,’ or ‘assistive-’ built environments including homes, schools, offices, museums, hospitals, libraries, stores, labs, prisons, warehouses, transport hubs, airports, parks, cities, etc. AmIE are often confused with (or described interchangeably as) ‘ubiquitous-,’ ‘pervasive computing,’ ‘mobile computing,’ or ‘spatial computing;’ as well as the Internet of Things (IoT) and building automation including HVAC automation. While AmIE comprise the aforementioned technologies, they specifically describe the paradigm in which these and other technologies (namely sensors, actuators, and processors) are outfitted in the built environment so that those places become sensitive and responsive to the presence, needs, wants, and preferences of their inhabitants. Given that AmIEs require an array of computing technologies to function, they are necessarily complex design-engineering systems that rely on some level of algorithmic taxonification and, as such, are vulnerable to perpetuating harm in so far as biases have been codified therein. This research engages this complexity and risk of harm perpetuation by utilizing Minor Theory (also ‘minor theory’) as a theoretical lens through which to begin to move towards values aligned taxonification of the vast AmIE design space. Through a series of sequential literature reviews, this work offers a jumping off point for the continued taxonification of the AmIE design space by 1) introducing and applying minor theory as a method for data analysis and by 2) proposing three thematic vectors to consider in codifying an AmIE a design space: attunement, embodiment, and anti-fragility.

In the early conceptual stage of design, novice designers and professionals experience different levels of fixation. Design tools and workflows can support designers to explore creative opportunities. Patent data tool for engineering design purposes is one example, using the design-by-analogy method to facilitate divergent ideation. This work will examine how designers can use the patent stimuli to explore both design problems and solutions and in what way.
The study applied a within-subject experiment with 6 participants. Each participant did brainstorming tasks with patent data tool and baseline tool. After the experiment, they gave feedbacks on the patent data tool. The concurrent think-aloud protocol, screen recording and idea sketches were used for explorative data analysis.
From the participants' self-report, they all gave positive feedbacks on the tool's value for acquiring new knowledge for divergent ideation. External raters clustered all the ideas by categories, following function-behavior-structure theory model. The overall fixation rate in the patent data condition did not bypass the baseline condition. From qualitative analysis on the ideation process, the patent data tool effectively eliminated knowledge-based fixation in the query stage. Its textual stimuli helped to release conceptual fixation if they were not exposed to solution visuals. The patent stimuli inspired participants with solution directly but did not provide problems consistently. The limitations of patent data tools are discussed. It points out the possibilities of using supplementary tools to better support exploration on problems, using analogies and thinking at an abstract level and overcoming conceptual fixation.
Lastly, future studies are needed to look into a combination of new tools to work with the patent data tool. Designers begin with understanding problem and user contexts from common sense knowledge tool. Afterwards, the patent data and biological-inspired tool can facilitate designers to ideate desired functions and related product behaviors. Ultimately, text to generative image tool can help conceptualize structural design. Future studies are needed to evaluate the effect of this new method.