At Zugvögel Festival 2026, our floor became a church - and what is a church without a confessional? So I built one: a wooden confessional box where festival visitors confess their sins to an AI priest, who teases the juicy details out of them, grants absolution, and then sends them onwards to the Holy Toilet - where their sins, anonymized and retold in an angel’s voice, are played back to whoever is enthroned there at the moment.
[Read More]
An interactive, networked spaceship you can actually board: players complete (or sabotage) missions across a dozen physical interfaces - buttons, joysticks, LED boards, sensors, etc - to either launch the ship or doom it, with everything driven by a Python/ESP/Unity stack. It’s modular, auto-resets regularly, and scales to festival crowds without manual babysitting.
[Read More]
First, there was the infamous PoolnudelschlagapparatPool Noodle Hitting Apparatus (PNHA). At Fusion Festival 2023, we had some pool noodles, some windshield wiper motors, and a wood construct to put a simple game into, so we did the only sensible thing to combine those things: You play Flappy Bird, and if you are too bad, you get hit with a pool noodle.
[Read More]
A universal relay box for interactive builds: Ethernet-controlled with an integrated PSU, switching 230V, 12V, 5V and dry relays via MQTT, OSC or Art-Net. I use it across projects like my Supertuxkart and the Palapa Spaceship to drive lights, fog, motors, and effects with simple, robust control.
[Read More]
In this (german-language) episode of Code for Thought, Carina Haupt (DLR) and I talk with host Peter Schmid about large language models in Research Software Engineering: where tools like Copilot genuinely help (routine tasks, tests, small scripts), and where real-world projects expose their limits (builds, requirements, missing project context). I share my current work on matching papers to repositories via embeddings at the project level rather than just functions - so models carry meaningful context across an entire codebase.
[Read More]
During my discontinued PhD at DLR, I developed and trained a graph neural network to combine code embeddings into repository-level representations and align them with scientific publications. Preliminary evaluation showed a promising improvement over pooling baselines, as a first step towards finding related implementations and their underlying assumptions across research fields.
[Read More]
VisioLab is a startup that creates automated self-checkout registers for canteens, where the meals on the tray are recognized on on-device via an iPad, allowing the customer to pay for their meals within seconds without requiring an employee.
[Read More]
The Siddata project developed a digital study assistant that helps students find relevant courses, learning resources, and study partners using personalized recommendations. I developed machine learning recommenders and worked extensively on its Django backend, turning a research prototype that previously took days to deploy into a containerized, tested, and easily deployable system. My masters thesis was also conducted in the scope of this project.
[Read More]
For this thesis, I created a Conceptual Space from Course Descriptions for explainable Recommendation, in a highly performant pipeline on the university-grid.
Sub-Pages:
[Read More]
After finally (and only for now) accepting that a reliable sound2light-bassline-detection is probably more trouble than it’s worth, I next needed a way for tap2bpm instead. This, however, required new Hardware, and thus, I developed this Button-Board.
[Read More]
As a student research assistant at the Red Hen Lab, I worked on a video-processing pipeline for automatic gesture analysis in large video corpora. My work involved person tracking and identification using face recognition, shot segmentation, and deploying the pipeline on an HPC cluster using Snakemake and Singularity.
[Read More]
As part of the practical course following the lecture The C++ Programming Language at the University of Osnabrück, we had two weeks to develop a complete game in a team of 11 students, including a level editor. The project involved working with libraries such as Box2D for physics and Boost, as well as the HDF5 data format. We decided to develop Bomberperson, a Bomberman-inspired multiplayer game with a few additional mechanics.
The Game
Bomberperson is a maze-based multiplayer game for 2-4 players, where the goal is simple: eliminate the other players and be the last one standing. Players can place bombs that explode after a short delay, sending explosions in multiple directions, destroying obstacles and potentially killing other players. Destroyed obstacles can drop power-ups, which can temporarily improve a player’s abilities or make their life more difficult.
[Read More]
A speech-controlled meditation assistant and sentiment tracker
[Read More]
An automated pipeline to split a bunch of scanned pages into single documents
[Read More]
Delving into the realm of self-driving cars for my bachelor’s thesis, I endeavored to optimize their tactical decisions, particularly speed, using cutting-edge deep neural networks and reinforcement learning in tensorflow within a transformed racing simulation. The resulting research platform, built on Unity, showcased the potential of real-time feedback and adaptability, offering valuable insights into the future integration of reinforcement learning in self-driving cars.
[Read More]
In this 3D-Browser-Game developed in a computergraphics-project using
THREE.js, the player is trapped in the infamous burning AVZ and must try to escape this inferno. The player controls the in-game character and must overcome obstacles, puzzles and traps to get to the bottom and out of the AVZ. Play here!
[Read More]