eprints (HSR Hochschule für Technik Rapperswil)
Not a member yet
1196 research outputs found
Sort by
Development of an LLM-first 2D Videogame
No full textTraditional role-playing video games (RPGs), where players assume the role of a character in a fictional
world, often rely on static dialogues and storylines. This can make replaying the game feel repetitive and
less engaging, as the story and conversations remain the same. In contrast, tabletop RPGs thrive on the
creativity of the human game master, who continuously invents new settings and stories and allows play-
ers to interact naturally with the world. With the rise of large language models (LLMs), new opportunities
are emerging for digital games. A key strength of modern LLMs is their ability to generate human-like
text in response to natural language input. Integrating LLMs into games enables live, context-sensitive
responses to player actions and dialogue, bringing some of the freedom and spontaneity of tabletop
RPGs into video games. This project aims to develop a 2D RPG with a top-down view that combines
traditional gameplay with the creative power of LLMs. At the start of each game, an LLM generates a
unique world. During gameplay, characters respond to the player in real time, meaning no playthrough is
the same. For developers, this moves focus away from scripting fixed content to designing systems and
prompts that enable the AI to take over parts of the storytelling.The game is developed in Unity using a 2D pixel art style and includes typical RPG elements such as story,
exploration, non- playable characters (NPC), battles, items and quests. Game content is partly generated
using LLMs and partly built from predefined assets. For example, character parts like heads, bodies and
clothes are defined in advance, and the LLM selects from these to create NPCs. Similarly, the map is
built from predefined 30x30 tiles depicting elements like forests, plains and villages, which the LLM as-
sembles into a complete world. The generation process starts with a prompt defining the map, followed
by prompts for quests, NPCs and items. NPC appearance and battle skills are generated in separate
prompts to keep outputs manageable. For simpler tasks like NPC visuals, skills and interactions, we use
OpenAI’s cost- efficient GPT-4o-mini. For tasks requiring more consistency and depth, such as map and
quest generation, we use the o3-mini model. We built a functional top-down 2D RPG with LLM-supported elements that generates new content on
each playthrough. Our key finding is that the quality of generated content depends heavily on the context
provided. Without sufficient details, quests and characters often become repetitive or inconsistent. For
example, NPCs might block quest progression, or items can appear in incorrect locations. While these is-
sues occasionally caused confusion, the eight people we selected to test our game generally responded
positively, appreciating the dynamic nature of the experience. A major challenge we encountered was
balancing the amount and complexity of prompts with system performance. Longer or more detailed
prompts improve output coherence but increase the risk of missing details. Splitting input into smaller
prompts helps but raises processing time. A promising approach is sending multiple prompts simulta-
neously for parallel processing, which could improve content accuracy and reduce wait times. Overall,
this project indicates that a thoughtful combination of AI and traditional techniques can open exciting
new possibilities for creating more dynamic and imaginative games in the future
Cloud-Optimized OSM GeoParquet Data Service for Switzerland and Beyond
No full textOpenStreetMap (OSM) is one of the most comprehensive openly licensed geospatial vector datasets, containing an estimated 60–90 million points of interest (POIs). While this is comparable to Overture Maps ~61 million POIs, OSM distinguishes itself through its data richness, openness, and crowdsourced quality assurance. However, its raw structure based on a graph of nodes, ways, and relations combined with a flexible tagging system, presents significant challenges for scalable querying and analysis.
This thesis presents a reproducible, open-source pipeline designed to transform country-scale OSM extracts, such as those from Geofabrik, into simplified, analysis-ready GeoParquet files. The files are aligned with Overture Maps Places and Divisions themes and converted into a tabular format optimized for geographic information systems. The solution is built on a modular Extract–Transform–Load (ETL) architecture using osm2pgsql with Lua scripts, PostgreSQL/PostGIS for schema alignment and spatial processing, and DuckDB with PyArrow for high-performance GeoParquet conversion.
Multiple spatial file partitioning strategies, including KDB Tree and S2, were evaluated to support efficient downstream interoperability and client-side filtering. The pipeline operates as a CI/CD enabled DataOps workflow, orchestrated via GitLab, containerized with Docker, and hosted on S3-compatible MinIO storage. A vandalism detection module prototype supports data quality by flagging anomalies in stable administrative names.
The result is Cadence Maps, a fully automated and publicly accessible data service for the D-A-CH-LI region, updated weekly and accompanied by release documentation. GeoParquet files can be queried directly via DuckDB or QGIS without requiring full downloads. For example, queries can filter specific features such as restaurants using hive-compatible S3 prefixes. Full dataset updates can be completed in under 24 hours, demonstrating the system's performance and scalability. This work establishes a reliable and extensible framework for delivering cloud-native geospatial data services with global potential
Content Analyzer
No full textTitel:
Website Advanced Content Analysis Tool
Untertitel:
Arbeitstyp:
Bachelorarbeit
Semester:
FS 2025
Studiengang:
BSc Informatik
Sprache:
Deutsch
Diplomanden:
Ali Al-Kubaisi, Simon Amberg
Referent:
Prof. Dr. Markus Stolze
Korreferent:
Reto Senn, bitforge AG, Zürich, ZH
Themengebiet:
Software Engineering
Einleitung:
Das Ziel dieser Bachelorarbeit war die Entwicklung einer Webapplikation `Content Analyzer` zur automatisierten Analyse von Webseiteninhalten.
Die Applikation soll Unternehmen dabei unterstützen, den Zustand ihrer Webseiten hinsichtlich Barrierefreiheit in Bezug auf Standards wie den den Web Content Accessibility Guidelines (WCAG) und Schreibstil effizient zu bewerten und zu verbessern.
Vorgehen / Technologien:
Im Zentrum der Arbeit stand die Umsetzung eines Web-Scrapers und eines modularen Systems, das open-source Analysetools über einen Plugin-Mechanismus integriert.
Der entwickelte Web-Scraper nutzt moderne Technologien, um Inhalte von Webseiten automatisiert zu erfassen und in einem Repository zu speichern.
Anhand dieses Repositorys wird dann die erfasste Webseitenstruktur mithilfe der eingebundenen Analysetools untersucht.
Die initiale Implementierung umfasste insbesondere die Analyse von Barrierefreiheit mithilfe des Tools Axe-Core sowie ergänzende Auswertungen hinsichtlich der WCAG-Richtlinien.
Zusätzlich wurde ein Schreibstil-Checker entwickelt, der regelbasierte und KI-gestützte Ansätze für grammatikalische Korrekturen und Textanalysen kombiniert.
Darüber hinaus bietet die Applikation Möglichkeiten zur Verwaltung und Annotation von Inhalten und Gruppierung, sowie Verwaltung von erkannten Problemen.
Ergebnis:
Besondere Herausforderungen beim Web-Scraping ergaben sich aus der Diversität an Technologien und Architekturen, die für moderne Webseiten eingesetzt werden, sowie der baumartigen Struktur von Webseiten.
Diese wurden durch detaillierte Analyse der Problemdomäne, sorgfältige Technologieauswahl, sowie einer iterativen Entwicklung erfolgreich bewältigt.
Das Ergebnis der Arbeit ist ein flexibles und wartbares System, das es ermöglicht, Webseiten auf grundlegende Richtlinie zu überprüfen. Über eine anschauliche Darstellung der Resultate können erkannten Problemen verwaltet und behoben werden.
In Zukunft soll das System als Grundlage dienen, auf der weitere Analysetools eingebunden werden können, um aufkommende Standards abzudecken und neue Fähigkeiten einzubinden.
Bild 1:
Analysis Dashboard: Plugin-Übersicht, Schreibstill, Content-Management, Coverage-Statistiken
Bild 2:
Modulare Architektur: Frontend, Webserver, Web-Scraper, Analyse-Services, MongoDB
Bild 3:
Accessibility Resultat mit WCAG-Analys
Semantic Clustering Toolbox
No full textThis thesis presents the design, development, and evaluation of a semantic clustering toolbox intended to support non-technical users in analyzing open-ended survey responses.
The motivation stems from a desire to expedite the labor- and time-intensive process of survey data analysis in research and evaluation contexts.
The toolbox allows users to upload survey data, perform semantic clustering, analyze sentiment, and export results through a simplified interface.
Developed using a Design Science Research methodology, it integrates embedding models for semantic representation, the K-means algorithm for clustering, dimensionality reduction for visualization, and language models for sentiment analysis.
A notable feature of the system is the inclusion of cluster stability visualizations, which help users interpret the consistency of clustering outcomes across multiple runs.
The artifact was evaluated through internal clustering metrics, user feedback and requirement validation using real-world survey data provided by the IFSAR research institute.
Results indicate that the toolbox effectively identifies dominant themes and supports exploratory analysis, while remaining accessible to non-technical users.
Despite its utility, the toolbox has limitations, including sensitivity to input quality and the inherent subjectivity of interpreting clusters without ground truth labels.
Nonetheless, the artifact fulfills its primary goal and offers a practical foundation for future enhancements and research.
Overall, this work contributes a practical and extensible tool for the semantic clustering of textual data
Finteam - A React Native App for Fiduciaries and their clients
No full textIntroduction
In Switzerland, tax filing is often perceived as a cumbersome process, which leads many individuals to delegate this responsibility to fiduciaries. However, this delegation introduces its own inefficiencies, particularly in the form of unstructured communication and incomplete documentation, which impair the ability of fiduciaries to work efficiently.
Approach and technology
This thesis presents the development of Finteam, a Minimum Viable Product (MVP) designed to streamline document exchange and communication between clients and fiduciaries. Starting from a basic prototype, the project improved the architecture, added core features, and developed a scalable backend to support future growth. The system consists of three main components: first, a mobile app (built with React Native and Expo) for clients to upload tax documents and interact with an AI assistant. Second, a web app (React + TypeScript) for fiduciaries to manage client submissions and provide feedback. Third, a modular backend (Express.js + TypeScript) that handles authentication, data storage, and an RAG-based tax assistant.
Conclusion
During this thesis, the team successfully developed an MVP for the Finteam application. The system replaces unstructured communication with a digital workflow that reduces administrative overhead on both sides. The mobile application enables clients to upload tax documents, while the web platform provides fiduciaries tye ability to review submissions and send feedback to the client. This approach eliminates the typical back-and-forth email exchanges that currently affect fiduciaries.
As a secondary but notable feature, the system incorporates an AI-powered tax assistant built on Retrieval-Augmented Generation (RAG) technology integrated into the mobile app. It combines vector- and keyword-based search with reranking and query rewriting techniques to improve response quality. An empirical evaluation of various configurations showed that reranking significantly enhanced the accuracy of responses. While this came with a latency trade-off of approximately 38% compared to vector-only retrieval, which served as a baseline. With this tradeoff, the overall user experience benefited from more relevant answers. Key evaluation criteria included helpfulness, correctness, clarity, and fluency.
Key limitations include the absence of a scalable fiduciary onboarding flow and restricted iOS deployment due to the lack of access to Apple's Developer Program. Future work should focus on automating onboarding, enabling administrative oversight, integrating document parsing to prefill tax forms, and at-rest encryption of user documents to enhance security
FutureNet Kantonsspital Graubünden (KSGR)
No full textThe Kantonsspital Graubünden (KSGR) is undertaking a comprehensive modernization of its aging network infrastructure through the FutureNet project. This initiative addresses the lifecycle replacement of network components and aims to implement a state-of-the-art, secure, and scalable network architecture. The existing classic three-tier architecture, segmented by Virtual Routing and Forwarding (VRF) and managed via a centralized firewall, is to be replaced by a Software-Defined Network (SDN) using Aruba’s EVPN-VXLAN fabric and NetConductor technology.
The thesis focuses on the design, integration, and validation of this new architecture, particularly at the campus LAN level. Key goals include the deployment of role-based access control, enhanced Network Access Control (NAC) with 802.1X authentication, and the implementation of macro- and micro-segmentation for improved security. A significant component of the work is the integration of the site Kreuzspital into the EVPN fabric via a distribution layer using Aruba’s VSX clustering approach.
After the new site was built and thoroughly tested, productive clients were successfully migrated to the new infrastructure. The entire project was carried out within the context of a live, productive hospital network, requiring minimal disruption to ongoing clinical operations and ensuring the highest standards of reliability and security
Standardisierung von Agenten zur Leistungserfassung in Klinikinformationssystemen
No full textIn Schweizer Spitälern werden medizinische Leistungen oft manuell erfasst. Das kostet Zeit und kann zu Fehlern führen. Die Software KISIM der CISTEC AG unterstützt zwar die digitale Leistungserfassung, doch die Agenten – Programme für die automatische Leistungsgenerierung – werden bisher individuell entwickelt und konfiguriert. Diese fehlende Standardisierung erhöht die Betriebskosten und verringert die Flexibilität, Agenten kundenzentriert weiterzuentwickeln.
Ziel dieser Masterarbeit ist die Entwicklung eines einheitlichen Backend-Frameworks für Agenten. Es stellt zentrale Funktionen wie Datenzugriffe, Protokollierung oder Fehlerbehandlung bereit und definiert eine klare Architektur zur Entwicklung neuer Agenten. Entwickler:innen können sich dadurch auf die jeweilige Fachlogik der jeweiligen Agenten konzentrieren, statt Basisfunktionen mehrfach umsetzen zu müssen.
Folglich vereinheitlicht das Framework auch die Codebasis und den Entwicklungsprozess: Neben dem Prototyp entstehen Demo-Agenten, die den Nutzen aufzeigen und eine Grundlage schaffen, die zur Effizienzsteigerung in der Entwicklung beitragen. Durch kohärente, getestete Agenten mit gemeinsamen, validierten Datentypen verringern sich die Komplexität sowie Wartungs- und Entwicklungszeit erheblich
Systematische Weiterentwicklung des Wortschatztrainers
No full textDiese Masterarbeit entstand in Zusammenarbeit mit dem Lehrmittelverlag Zürich (LMVZ) und zielte darauf ab, den digitalen Wortschatztrainer des Französisch-Lehrmittels «disdonc!» zu überarbeiten und nutzerfreundlicher zu gestalten. Grundlage war das Collaborative UX Design, das nach einer Evaluation verschiedener Vorgehensmodelle aufgrund seines kollaborativen Ansatzes gewählt wurde.
Zu Beginn wurde eine Planung für das Projekt erstellt. Verschiedene Vorgehensmodelle wurden evaluiert, um das Passende für dieses Projekt zu finden. Das Projektteam entschied sich für das Model Collaborative UX Design. Dazu wurde eine Risikoanalyse und ein Stakeholdermanagement entwickelt, um die beteiligten Akteure rund um den Wortschatztrainer zu berücksichtigen. Der LMVZ war mit zwei Vertreter:innen aktiv eingebunden, die regelmässig über Fortschritte informiert und in wichtige Entscheidungen einbezogen wurden.
Die entwickelte Lösung des Wortschatztrainers (WST) basiert auf in qualitativen Untersuchungen erhobenen Daten, die nach einem klar strukturierten Vorgehen erarbeitet und validiert wurden. Dabei entstanden essenzielle Artefakte wie die Primäre Persona und die User Journey, welche die Grundlage für ein präzises Verständnis der Nutzerbedürfnisse und der Problemstellung bilden. Im Problem Statement Map wurden bestehende Herausforderungen sowie neue Zielsetzungen dokumentiert.
Die Analyse der bestehenden Lösung erfolgte auf Basis definierter Kriterien, wobei gewonnene Erkenntnisse gezielt in die neue Lösung integriert wurden. Die Ausrichtung an den fünf Dimensionen von W. Quesenbery erlaubte eine gezielte Fokussierung auf relevante UX-Metriken, wodurch die Ergebnisse zukünftig messbar sind. Diese fundierte Grundlage sorgte dafür, dass während der Ideations- und Konzeptionsphase ein klarer Produktfokus vorhanden war, gleichzeitig jedoch genügend Raum für kreative und innovative Ansätze blieb.
Das Szenario und die User-Story Map halfen, die zentralen Ziele weiter zu konkretisieren und Interaktions- sowie Funktionspunkte gezielt zu priorisieren. Auf dieser Basis wurde ein Prototyp entwickelt, der durch mehrere Iterationen getestet und optimiert wurde. Nach Abschluss des Testings wurde eine Priorisierung der erarbeiteten Features durchgeführt, wobei eine klare UX-Matrix sowie der Entwicklungsaufwand berücksichtigt wurde.
Die finale Lösung zeichnet sich durch eine verbesserte Benutzerfreundlichkeit, didaktische Qualität und eine stärkere Ausrichtung auf die Bedürfnisse der Schüler:innen aus. Während dem Projekt wurden dazu elf Schüler:innen befragt und zwei Testingrunden mit weiteren elf Schüler:innen durchgeführt. Dabei hat das Projektteam gelernt, wie wichtig ein angenehmes Umfeld für junge Testpersonen ist, und das Verständnis zum Thema Lernverhalten wie auch zur Didaktik konnte vertieft werden
Towards Greener Software: Measuring Performance and Energy Efficiency of Enterprise Applications
No full textPerformance is an important aspect of software quality in most software applications. In recent years, the topic of energy efficiency has become increasingly important. Enterprise applications running in the cloud receive particular attention. Poor performance and low energy efficiency lead to high operating costs and a negative impact on the environment. Recognizing the significance of performance and energy efficiency as software quality attributes and how to measure them is crucial. This research project aims to investigate how two types of software quality attributes, performance and resource and energy efficiency, are measured according to the state of the art and the practice today; differences are identified and analysed. Following an empirical approach, an existing application test setup and its measurement results (publicly available in the “Growing Green Software” blog) are first reproduced and then compared with the behaviour of a second sample application. The two respective sample applications are open source-projects leveraging Java and Spring Boot; one of them comes as a set of microservices. Tools such as JMeter and JoularJX, configurations, and metrics across different test environments and enterprise applications were experimented with. Contemporary software engineering practices such as Domain-Driven Design and UML were used to analyse and document the software architectures of the selected applications. The measurements confirmed that the performance and energy consumption of the application are significantly influenced by external factors such as hardware, operating system, and implementation details. They show that the relative distribution of energy consumption is comparable across different test environments and enterprise applications. Furthermore, the results suggest an inverse correlation between performance and energy efficiency when different hardware is compared. They also indicate a strong correlation when the same hardware is used but implementation details vary. Future research could explore the energy efficiency of cloud-native applications and cloud infrastructure
A Command-Line Tool for Managing Recurring Architectural Decisions: Design, Implementation, and Empirical Evaluation
No full textThis thesis presents the design, implementation, and evaluation of ADG (Architectural Decision Guidance), a command-line tool for managing recurring architectural decisions. Based on previous conceptual work and a Clean Architecture decision handbook, ADG offers a structured, metadata-driven approach to managing the lifecycle of Architectural Decision Records (ADRs). Implemented in Go and aligned with Clean Architecture principles, ADG supports modular decision models using editable Markdown files, nested links, and guidance templates.
ADG facilitates the creation, tagging, linking, validation, and reuse of ADRs through 18 command-line interface (CLI) commands, promoting model-driven knowledge reuse across projects. One of its key features is support for guidance modeling, which allows curated decision templates to be instantiated and merged across contexts.
An empirical evaluation involving a software architect and a graduate student demonstrated the usability and practical relevance of the tool in realistic scenarios. ADG distinguishes between three roles: Knowledge Engineers, Model Tailors, and Architects/Developers. These roles enable collaboration through structured workflows. The result is an extensible CLI prototype that advances the practical tooling landscape for decision-centric architecture modeling