Flashcard app that I started building out of frustration with other apps. The two goals are ease of flashcard creation using a templating system and a flexible learning algorithm that does not limit the user.

• Simple Python backend architecture
• GUI built using web technologies (HTML, CSS, JS)
• An open database format using sqlite3
• Compilation into a Windows and Android app

Lumi RPN is a desktop Reverse Polish Notation (RPN) calculator for Windows, written in Python using a customtkinter GUI.

• Keyboard input
• Math functions as buttons
• Use of a stack-based (RPN) input method
• SI unit tracker
• Math and physical constants

A graphics demo about rendering 3D objects and moving freely in real-time written using C++ and the OpenGL API.

• Extensible project structure using OOP
• OpenGL rendering pipeline
• 3D mesh loader for .obj files
• Level creation via .json files

A modular JavaScript/WebGL game engine built using Three.js, focused on isometric rendering and tile-based level creation using JSON files.

• Player movement implemented
• Tile-based levels using Texture atlases
• World & level loader from .json files
• Entity & object model
• UI System (planned)

Scripts to digitize large amounts of scans automatically using OCR. Using this, a whole book can be turned into structured text files to create a PDF file.

• Character recognition using Tesseract's pre-trained OCR models and easyocr
• Spell checking
• Recognizes text blocks and paragraphs automatically
• Automatic PDF → Image file → Text pipeline

Procedurally generated snowflakes using convolution kernels. Using a random seed beautifully looking snowflake images are created. One possible adaptation of this concept could be to randomly generate intricate maps for video game levels from random seeds.

• Iteratively applied convolution is similar to cellular automata
• Hexagonal symmetry is achived later using a skew transform

By applying the concept of a five dimensional framework of Conformal Geometric Algebra (CGA) from physics, I tried to generate some scene renders using raytracing methods. Methods for this are described in Hadfield et al , where they also introduce parametric volumes that could be used in graphic design and geometry-heavy applications.

• Raytracing and lighting using Blinn-Phong shading
• Implementation of parameterized surfaces
• Not a single linear system of equations used, but Clifford algebras to solve for ray paths
• Reflective surfaces

Tool for inserting logos/branding and simple images into a QR code, making it become visible especially from for away. The QR-Code generator uses Error Diffusion Dithering to insert an image directly into the scannable module matrix. Available at qixel.eu.

• Manually mark the image foreground object
• Automatic foreground detection using GrabCut
• Image dithering and insertion into a QR code
• QR code is embedded into the dithered image, so the readablity is less impacted

University project: Creation of an ML model and training of a bot that is capable of playing the popular online game GeoGuessr. Geoguessr is an online game that challenges players to guess the location of a randomly selected place in the world based on a Google Street View image.

• Geoguessr's point system already provides a metric to optimize
• Dataset sourced from Googles Streetview API
• Design of Convolutional Neural Networks
• Using a cloud computing service to train
• The country classification model did not yield satisfactory results
• An inception model trained on image tiles showed the most promising results