Software Engineering

This Master’s thesis aims to enhance the educational platform Artemis by implementing automatic formative feedback for students and improving semi-automatic assessment for tutors. Building on the foundational work of Athena and CoFee, which use NLP and LLMs for text-based feedback, this project focuses on refining these techniques for greater accuracy, reliability, and consistency.

The initial phase will prioritize developing immediate feedback mechanisms to support students before deadlines. Following this, we will explore advanced LLM techniques like RAG, CoT prompting, self-consistency, and fine-tuning models like Llama 3 and GPT-4 with historical feedback data. Each iteration will be evaluated for accuracy, efficiency, and educational impact to ensure continuous improvement. By adopting an agile approach, we aim to deliver automated, personalized feedback that enhances student learning and reduces tutors’ workload, creating a more efficient and supportive educational environment.

In this thesis, we develop a gamification-based system to enhance the code review process by increasing participation and improving code quality. By integrating gamification elements such as points, levels, badges, and leaderboards into existing code review platforms, we aim to create a more engaging and motivating environment for developers. Our system provides real-time feedback through a user-friendly web interface, and seamlessly integrates with popular development tools to support continuous workflows. The primary contribution is a versatile framework adaptable to various code review systems, fostering a culture of continuous improvement and collaboration among developers. To evaluate the system’s effectiveness, we will conduct a comprehensive user study involving multiple development teams. This study will use both quantitative metrics, such as participation rates and review quality scores, as well as qualitative feedback from user surveys and interviews. The results will provide detailed insights into the impact of gamification on code review practices, demonstrating its potential to enhance software development processes and other collaborative environments.

In modern development workflows, the seamless integration of UML diagrams into documentation, issues, and pull requests on Git hosting services is crucial for clarity and collaboration. We can significantly enhance Apollon, a web-based UML editor, to meet these needs. This project aims to implement efficient embedding capabilities, allowing developers to embed their UML diagrams with live updates in issues, pull requests, and more. When users make changes to diagrams in Apollon, the system will automatically reflect these changes in their embedded versions, eliminating the need for manual updates. Additionally, we will introduce version management, which will enable users to capture new versions, preview changes, and revert to previous versions. Furthermore, a new structured and readable textual format for diagrams, integrated with VSCode, will facilitate developers in versioning their UML diagrams with Git for seamless documentation integration. These enhancements will improve Apollon’s usability, encouraging its adoption in development workflows.

Abstract TBA

Competency-based education gives students transparency and flexibility in their learning process. The usage can enhance motivation, especially in university settings where instructors address students with diverse backgrounds and prior knowledge together. Therefore, Artemis contains the adaptive learning system Atlas to support university teaching. To provide the best experience possible to students, it is essential to understand how students currently interact with the system integrated into Artemis and its impacts on their learning process.

The growing use of Large Language Models (LLMs) presents challenges in selecting the most suitable model for a given prompt. Each LLM has distinct strengths and limitations, making model selection a crucial factor in optimizing response quality, processing time and computational efficiency. Currently, prompts are often processed without prior classification, leading to one overall model choice that increases costs or reduces output quality.

This project introduces Logos (Leveraging Orchestrated Governance for Optimized Sustainability), a system for centralized LLM configuration, request routing, and efficient resource utilization.

Competency-based education(CBE) enables flexible, skill-centered learning by allowing students to progress based on mastery rather than fixed schedules. Atlas already utilizes machine learning techniques to support educators; nevertheless, further improvements are required to accelerate adoption and improve quality.

This thesis proposes an interactive AI-powered agent that assists instructors in creating, refining, and maintaining competency networks through natural language. Rather than relying on static, one-shot inputs, the agent engages in a dynamic conversation—asking clarifying questions, gathering relevant context, and presenting interactive proposals. This interaction model is designed to reduce redundant AI calls, avoid naive prompting, and ensure that the resulting competency networks align closely with course objectives.

TUM lacks a unified, user-friendly platform for managing doctoral applications. Fragmented systems lead to inconsistent information, limited transparency, and accessibility barriers. Unclear navigation and outdated interfaces hinder applicants, while manual tasks burden faculty and staff. A centralized, inclusive solution is needed.

This thesis designs, implements and evaluates a board game using Ozobots to address gender stereotypes in STEM education among elementary students. It reviews gender-sensitive strategies, assesses robotics tools for inclusivity, and develops an engaging board game to promote equal participation and interest in STEM fields for both boys and girls.

Pupils playtest two board games that are identical in structure but differ in theme. Each version addresses a distinct gender stereotype. The process records engagement, collaboration, and individual perceptions through structured surveys and classroom observations. The collected data will provide insights into pupils’ confidence and participation levels, guiding refinements to enhance inclusivity.