Simon Kaier

Snippeter.net is a platform where you can store or share your most exciting code snippets.

It has community features such as sharing snippets, commenting on them, and leaving reactions.

The focus of this project is JavaScript and you can run JS code directly from the snippet in the browser and see the results.

Snippeter.net is built using Angular in the front end and Hasura (GraphQL to PostgreSQL Engine) in the back end. It's hosted on AWS RDS and EC2.

It's built with high snappiness, responsiveness. fast response times in mind and fully open-sourced: https://github.com/sn-kaier/snippet-manager

A Java Swing application to visualize and explore the Mandelbrot set.
The main focus was smooth color grading using splines and deep zoom using DoubleDoubles (a technique where two 64 bit doubles are combined to represent a more accurate floating-point number).

Mandelbrot renderings are computationally heavy.

To enable the user to zoom into the fractal smoothly without having to wait 10+ seconds for every frame to render some performance improvements were made:
1: Rendering on all available cores.
2: Incremental rendering (splitting the image into individually rendered chunks).
3: Determination of the "most interesting" areas/chunks using the previous frame to be able to render them first.

Other approaches for performance have been leveraging the GPU using OpenCL and distributed computing on multiple machines.

Lessons learned: Consumer GPUs are optimized for 32-bit floats. Their 64-bit floating-point operation performance is usually not significantly higher than the one of a CPU.

For my own room, I put ~18 meters of RGB LED strips in the shadow gaps around the whole room and soldered them + wired them to the Teensy.

On a Teensy controller (comparable with Arduino but more powerful), I programmed different light modes in C (running light, wavelike evaluations with several colors, etc.).

The Teensy was hooked up to a Raspberry Pi via USB which was running a Node.js server to communicate with a mobile web application over the wifi.

The web application was built using Vue.js and hat basic functionality to control the light: brightness, four different light modes, color choice, and an on/off switch.

The overall setup looked like this:
LED strips --> Teens micro controller/C --> Raspberry Pi/Node.js --> Smartphone/Vue.js

The practical part of my bachelor thesis at the University of Stuttgart was comparably large.

The task was the extraction of names of persons from large data sets consisting of twitter messages using natural language processing tools and the subsequent dynamic extraction, Incremental representation of the data over a time axis.

I developed a back end that would process the text data and expose the results in a useful way through a REST endpoint.

Key technologies included Kotlin (JVM), Stanford NLP (to extract names), and MongoDB for storing and retrieving data using aggregations.

For the visualization, I implemented a web application using Angular and D3.js.

Final Grade: 1.0 (Highest marks)

A project at VIS Institute (University of Stuttgart) as one of three other working students. The goal was to be able to interactively explore small to medium-sized graphs in a web application.

I worked for three months on that project and during this time my responsibilities were:
1: Identifying and containing performance bottlenecks.
2: Implement an adjacency matrix visualization.
3: Algorithmically recognize and display groups (closely networked nodes).

As the third semester project, two fellow students and I had to compete against 30 other teams to build the best indoor navigation app for the computer science building.

My responsibilities in the project were:
1: Creating a virtual map of the building.
2: Building an application to model a graph for the floors and stairs using Java Swing.
3: Implementing a suitable pathfinding algorithm using Dijkstra.

Results:
We were the winning team and gained a prize of over 150€ per person.
We stood out because we used Wifi triangulation to track the user's position during navigation and we had a 3D live view of the building similar to AR.