CS MSc Thesis Presentation Day March 19 2026

Thursday March 19 is a day for coordinated master thesis presentations in Computer Science at Lund University, Faculty of Engineering. Three theses will be presented.

You will find information about how to follow along under each presentation. The presentations will take place in E:2116. A preliminary schedule follows.

Note to potential opponents: Register as an opponent to the presentation of your choice by sending an email to the examiner for that presentation (firstname [dot] lastname [at] cs [dot] lth [dot] se). Do not forget to specify the presentation you register for! Note that the number of opponents may be limited (often to two), so you might be forced to choose another presentation if you register too late. Registrations are individual, just as the oppositions are! More instructions for opponents are found here on the LTH thesis project page.

10:00-11:00 in E:2116 (N.B. No more opponents for this presentation)

• Presenters: Karl Alemo, Fredrik Orheim
• Title: Efficient Fuzzing of Web APIs in Embedded Environments
• Examiner: Per Runeson
• Supervisors: Alma Orucevic-Alagic (LTH), Victoria Vucic (Axis Communications)

Background: Test generation by fuzzing can reduce testing effort for web APIs in embedded environments. However, adopting a fuzzing tool in a large company context raises concerns regarding efficiency, usability, software security and legality. Aim: This project aims to identify key fuzzing tool aspects required for integration in a large industry setting, while evaluating open source alternatives. Method: A Design Science Research approach was applied to elicit key aspects through a literature study, semi-structured interviews and a focus group. Open source alternatives were selected, and their performance was statistically analysed across company-specific APIs. Results: A proof-of-concept solution, using EvoMaster, aimed to comply with the identified required tool properties. Conclusion: While the majority of fuzzers are not currently mature enough for commercial adoption, there are tools that fulfil the identified key fuzzing tool aspects and, upon further refinement and evaluation, fuzzing could provide value in a large industry setting.

Link to popular science summary (PDF, 118 kB, ny flik)

14:15-15:00 in E:2116 (N.B. No more opponents for this presentation)

• Presenters: Axel Nåsell, Harry Norrman
• Title: Creating an Onboarding Tool for Technical Integration Using an LLM Approach
• Examiner: Elizabeth Bjarnason
• Supervisors: Sule Tekkesinoglu (LTH)

Technical integration of new developers in the onboarding process is an important aspect for companies to grow their workforce efficiently. This is a difficult process that often leaves new developers confused and demands resources from experienced developers to get them started. This thesis explores the possibility of creating an LLM-based assistant that can expedite the onboarding process. Through the design science research methodology, we identified current issues related to technical integration within the onboarding process, implemented an agentic AI solution called Bosch Onboarding Buddy (Bob) and evaluated its performance through user studies. In relation to current challenges, we found that the biggest issue is finding the most relevant information related to a task assignment within a large documentation database. We implemented Bob using Retrieval-Augmented Generation (RAG) to enable real-time information fetching to solve this problem. Our evaluations showed promise in the ability to find and summarize relevant information. While Bob cannot replace the value of receiving assistance from senior developers, it complements the existing technical integration tools and strategies.

Link to popular science summary to be added

15:15-16:00 in E:2116 (N.B. No more opponents for this presentation)

• Presenter: Edwin Gustafsson
• Title: Animated Neural Billboarding for Real-Time Rendering
• Examiner: Michael Doggett
• Supervisors: Rikard Olajos (LTH)

In real-time computer graphics, billboards are a frequently used way to save on computation time by simplifying an object down to a flat image. Billboarding for video game rendering has been used since the first 3D games and continues its popularity today, with a common use case being trees. Often these billboards are animated to show movement in the object, like a tree swaying in wind. Billboards come with drawbacks, such as not being able to realistically relight the object with self shadowing and complex light interaction. Previous work has proposed neural shading as a solution, running a small neural network in the fragment shader to encode light interaction, though only on a static billboard. In this work, we continue the proposed neural shading solution by adding the ability to animate the billboard texture. We present an ablative study of four model architectures along with image evaluation and performance metrics.

Link to popular science summary to be added