Hi, I am Aris and I study MSc in Human-Computer Interaction

Approaching human-centered design with an analytical, data-driven mindset

My Journey

My Story

How it started

I've always been curious about how things work, technology felt like magic to me, and wondering about what is underneath is something that never stopped being exciting.

That curiosity pulled me toward math and Computer Science. I got my Bachelor's in Computer Science and Engineering and I spent then I spent 1.5 years working as a software developer. But the more I built, the more I realized my curiosity wasn't just about the technology, it was also about the people using it.

How it is going

That realization is what led me to Utrecht, to pursue a master's in Human-Computer Interaction. I wanted to keep the analytical thinking I'd built as a developer, but point it at a different question: not just how to build something well, but how to build something that actually serves the people using it.

Studying HCI lets me combine both sides of how I think, bringing the same rigor I apply to code and data to understanding people, and using that to design solutions that put human needs first.

My aspirations

I envision myself as someone who can design and build digital products that put human needs first — and this master's is my path there.

Along the way, I'm deliberately sharpening my data-oriented skills too, through electives in research methods and statistical analysis, so I can back up design decisions with evidence, not just intuition.

School Projects

Projects

Mouse Interaction Experiment

Task

The computer mouse remains one of the most used input devices, so small differences in pointing speed and accuracy can accumulate into meaningful effects over time. Most operating systems enable 'mouse acceleration' by default, a non-linear mapping where cursor displacement depends on movement velocity, not just distance. This can aid fine control at low speeds, but it also makes movement less predictable, which is why gaming communities often disable it, mostly on anecdotal grounds. We wanted to test this empirically, over a longer timeframe than prior single-session studies. Research question: How does disabling mouse acceleration influence pointing speed and movement accuracy in everyday computer use?

Action

As a team of four, we built a custom web application in Svelte to run a battery of clicking, slider, and dragging tasks based on Fitts' Law, tracking cursor movement at 60Hz and logging it to a database for analysis. My part included building the dragging task, refactoring parts of the codebase, and — together with two teammates — handling the statistical data preprocessing and writing the methods and research gap sections of the report.

Mouse clicking task

Mouse dragging task

Result

Across a two-week study with 19 participants, the group who disabled mouse acceleration showed significantly steeper improvements in pointing performance (Fitts' throughput) across all three task types compared to the group who kept it enabled, along with smoother, more efficient movements over time. The effects were statistically significant but small in size — evidence that a stable, linear mapping between hand and cursor movement helps users build more consistent motor skills, even if the everyday impact is modest.

Reflection

This project taught me how to run a real longitudinal experiment end-to-end: designing tasks around an established theory (Fitts' Law), collecting weeks of behavioral data from real participants, and analyzing it with linear mixed-effects models rather than a single-session comparison. It also showed me how much noise remote, real-world data collection introduces, and how to design an analysis that's still robust to that.

Key Learning: Longitudinal experiment design, Fitts' Law-based task design, Svelte, statistical analysis with linear mixed-effects models in Python.

Adaptive Recipe Recommender System

Task

Deciding what to cook is often overwhelming, and while most recipe platforms focus heavily on taste and ingredients, they largely ignore contextual factors like time, skill, and available equipment that determine whether someone actually cooks a recipe. Food choice is also heavily driven by visual cues, yet existing platforms typically show only an image, title, and rating on their overview cards. We wanted to design a recommender that surfaces cooking experience directly in the interface, not just the algorithm. Research question: How do visualizations of cooking experience metrics influence user engagement and the accuracy of personalized recommendations in adaptive recipe systems?

Action

As a team of four, we designed a recipe recommender combining a hybrid content-based and collaborative-filtering algorithm with three cooking-experience metrics — difficulty, equipment overlap, and a Prep & Cook / Tend / Rest time breakdown — visualized directly on recipe cards through colour-coded badges and a donut chart. We also proposed an onboarding flow to ease the cold-start problem, a two-page recipe card that separates decision-making from execution, and a post-cooking questionnaire that keeps the user model adaptive over time.

Carousel Layout of Recipes

Recommended Recipe Card Design with badges

Completion Recipe Screen

Result

As a research proposal, the system wasn't built and tested end-to-end, so instead we designed a layered evaluation plan to validate it in stages: verifying the accuracy of interaction logging, an expert heuristic review of our data-interpretation rules, an offline test of the hybrid algorithm against the Food.com dataset (180K+ recipes), and a planned 3+ week between-subjects study comparing five badge conditions (none, each badge alone, and all three together) to measure whether the full system improves cook-through rate, decision speed, and satisfaction.

Reflection

This project taught me how to translate an abstract research question into a concrete, testable system design — defining specific metrics, justifying algorithmic choices like a weighted content-based/collaborative-filtering blend, and structuring an evaluation across separate layers (data, interpretation, algorithm, whole system) instead of testing everything at once. Incorporating peer and lecturer feedback also pushed us to simplify the design more than once, for example replacing a cluttered multi-carousel overview with a single scrollable list.

Key Learning: Adaptive system design, hybrid recommender algorithms, layered evaluation methodology, UI design for reducing cognitive load.

Data Visualization Methods

Task

Data visualization is becoming increasingly critical as the volume of available data grows, especially for representing complex, high-dimensional, and relational structures like graphs and networks. Poor visualization choices can distort the very insights they're meant to reveal, so choosing the right layout and technique for a given graph's structure matters. For this project, we set out to build a graph-visualization system from scratch, implementing and comparing multiple drawing algorithms to understand which techniques work best for which kinds of network structure.

Action

As a team of three, we built a full graph-visualization toolkit in Python (NetworkX, Matplotlib), implementing and comparing several layout algorithms: random and circular layouts, a custom layered tree layout for BFS/DFS spanning trees, the Fruchterman-Reingold force-directed algorithm, a hierarchical Sugiyama-style layered layout with the Eades heuristic and barycenter/median crossing minimization, and dimensionality-reduction-based projections (classical MDS, force-directed MDS, ISOMAP). We also implemented multilayer graph visualization with Kernighan-Lin cluster partitioning and force-directed edge bundling to reduce clutter between clusters.

ISOMAP vs. MDS projections of the same graph

Result

No single layout worked universally: random layouts conveyed no structure, circular layouts only made sense for near-complete graphs, and BFS trees became illegible on dense networks while DFS trees fared better on sparse graphs but degraded into unreadable diagonal chains on dense ones. Our force-directed implementation matched NetworkX's built-in spring layout in quality, and for the hierarchical layout, the Eades heuristic combined with edge curvature produced far more balanced and readable results than a trivial edge-reversal heuristic. Edge bundling worked well for clearly separated clusters but collapsed into an unreadable band when automatic partitioning produced too many inter-cluster edges — a visualization technique is only as good as the structure it's given to work with.

Reflection

This project taught me how many different ways there are to represent the same graph, and how the 'right' choice depends entirely on the structure and density of the data — a lesson that only became clear by implementing the algorithms myself rather than calling a library function. Translating academic papers (Fruchterman-Reingold, the Sugiyama framework, force-directed edge bundling) into working code also deepened my understanding of algorithmic complexity and the tradeoffs between visual clarity, computational cost, and correctness.

Key Learning: Graph drawing algorithms (force-directed, hierarchical/Sugiyama, MDS/ISOMAP), Python (NetworkX, Matplotlib), algorithmic complexity analysis, data visualization design principles.

Theory in Practice

Side Projects

Table Tennis Student Association Website Redesign

Table tennis has been my hobby since childhood, I used to play competitively before stepping away during my bachelor's. When I moved to Utrecht for my master's, I found the University Table Tennis Association (UST) and got to play again. Being part of UST also gave me the chance to put my computer science background to use: I redesigned the UST website, first mocking it up in Figma and then building it in WordPress together with a friend, so future members could manage it themselves.

Website redesign mockup

Key Learning: Figma, WordPress, Web Design

Visit UST Site →

Ferdinandos Hair Product Website

Ferdinandos, the owner of a barbershop in my hometown of Ioannina, asked me to help launch a premium hair product he's developing. I'm currently designing and building the e-commerce website for it: I started by designing the full experience in Figma, and I'm now implementing it in Next.js and TypeScript as an online store where customers across Greece can purchase and have the product shipped to them. It's a chance to take a real client's product from design concept to a production-ready storefront, end to end.

Coming Soon Landing Page

Key Learning: Figma, Next.js, TypeScript, E-commerce Web Design

Visit Ferdinandos Site →

Reflection

Study Plan

Goals

I didn't start this master's with a course-by-course plan. Being a developer felt limiting: I was shipping features, but I wanted to understand the humans those features were built for. I've always liked considering human needs when creating something, and I enjoy researching what people actually want and understanding their behavior. At the same time, I never lost my love for building and implementing. So the goal of this master's was to gain a theoretical understanding of humans, how they think, behave, and make decisions, and combine that with my technical expertise to create technology that genuinely serves people, not just features that ship.

Achievements

That blend has already started to take shape in my coursework. In Adaptive Interactive Systems, I helped design a recipe recommender that treats human context, such as time, skill, and available equipment, as seriously as taste, translating research on food decision-making into concrete interface metrics. In Data Visualization, I got to work at the more analytical end, implementing graph-drawing algorithms from scratch to understand how structure itself can be made legible to people. And in an empirical research project on mouse acceleration, I ran a full longitudinal study, from experiment design to statistical analysis, to test a claim about human motor learning that had previously only been anecdotal. Across all three, the pattern is the same: using rigorous, technical methods in service of a genuinely human question.

Looking Ahead

Looking ahead, I want to move into research and development roles that let me work the full cycle myself: understand what people actually need through research, then design and build the thing that addresses it, rather than handing that boundary off to someone else. The rest of this master's, including my thesis, is where I want to keep sharpening that loop: research methods sharp enough to trust, and technical skills strong enough to act on what that research reveals.

Expertise

Skills & Knowledge

User Research

Designing studies that actually answer the question you're asking, from framing a research question and choosing a design, to running it with real participants. Practiced across a longitudinal experiment on mouse acceleration and a layered evaluation plan for an adaptive recipe recommender.

Quantitative Analysis

Analyzing behavioral data with statistical rigor, comfortable going beyond a simple t-test. Ran linear mixed-effects models with multiple-comparison correction in Python to test whether disabling mouse acceleration improved pointing performance over time.

Qualitative Analysis

Interpreting non-numerical signals, from expert heuristic evaluations to open-ended feedback, to judge whether a design decision actually holds up. Used to validate interaction-interpretation rules for a recipe recommender's user model.

Implementation

Turning a design or research idea into a working product, not just a mockup. Built full task batteries, dashboards, and client websites end-to-end in Svelte, Next.js, and TypeScript.

Recommender System Design

Designing recommendation algorithms that balance accuracy with real user context. Proposed a hybrid content-based and collaborative-filtering model that factors in cooking skill, time, and equipment, not just taste.