Selected Work and Projects
ContextFlow
Categories:
OOP Architecture
Package
Infrastructure
Tests
Documentation
LLM Integration
Automation
Asynchronous
A .NET library for interacting with LLMs (Large Language Models). It provides abstractions for individual modules to work together through dependency injection. With it, you can create more complex prompt chains/pipelines as it simplifies prompt templating and prompt chaining, among other things, through a project-wide fluent interface. The entire project also includes asynchronous counterparts and is tested with 55 NUnit tests. Furthermore, attention was given to developing with a cyclomatic complexity (<6 for methods), class coupling (<10 for methods), and a Maintainability Index of 82. You can find the documentation here.
Finsights
Categories:
Product
Automation
Webapp
Branding
Logo Design
API
A Newsletter as a Service solution for all kinds of investors to keep in touch with their investments. Users can choose when, and about what they are notified about. They will then receive automatic performance updates for all selected stocks within the chosen time period. Microsoft is currently blocking SMTP access, which is preventing emails from being sent.
GPTVault
Categories:
LLM Integration
Proof of Concept
Knowledge Management
Automation
What you are seeing is a conceptual graph generated from GPTVault. GPTVault is a proof-of-concept tool that uses generative models to accumulate knowledge in a way that makes it usable. It works by querying a large language model (LLM) about a starting topic. The LLM will respond with a brief description of the concept, a list of related concepts, and a list of subconcepts. This information is then saved in a knowledge base, which can be explored and visualized in a variety of ways. GPTVault can be used for a variety of purposes, including concept exploration, personal knowledge management, and content creation. It is a tool that can help you to learn more about or organize a concept.
Embyte
Categories:
Webapp
Caching
Algorithm Development
UI/UX-Design
Minimalistic Design
The open-source project that allows you to create customizable embeds displaying website information and embed them using iFrames. It was inspired by Discord's embeds. This project caches requests in a PostgreSQL database and utilizes a self-developed caching algorithm.
Time-Series Forecasting of PV-Growth
Categories:
Machine Learning
ML-Model Validation
Data Science
Data Extraction & Processing
Data Visualization
Time Series Forecast
Used a series of mutlivariate time series models from the package Darts to forecast the energy production from PV-systems in Germany using 8 historic and 1 predicted metrics. Optimizing mode architecture (N-Beats), evaluating and using the forecast in a research-paper about the consequences of the national growth in pv-systems.
Mosaic: Link as bio for projects
Categories:
Product
Minimalistic Design
Freeware
Backend
Frontend
Database
In a team of 2 people, I developed a web app that enables the 'Link as Bio' concept for projects. In this project, I primarily focused on developing the backend and the database. Users can sign up and add projects to their account, which can then be linked.
Reader
Categories:
Webapp
UI/UX-Design
Minimalistic Design
Freeware
A highly customizable reader built as a direct alternative to paid speed-readers. Concepts such as chunking, pacing and highlighting are built into this reader. Texts can be imported from files or the clipboard. All data is stored in the website's local storage.
Instadata
Categories:
Data Extraction & Processing
API
Scraping
Automation
A large instagram scraper built with Django that uses an API to communicate between user and program. The Scraper can be scaled seaminglessly and uses proxies. I built it to gather data for further Data-Science projects, so I used the Nominatim Geolocator and Regex to extract and order as much data as possible.
Soccer Prediction Model & Evaluation
Categories:
Machine Learning
Monte Carlo Simulation
ML-Model Validation
Data Science
Data Visualization
Data Extraction & Processing
This project explored predicting football match goals using machine learning. The goal was to test a goal-count-based regression model against the default choice of a win/loss/draw classification model. Despite achieving some success in predicting match outcomes, limitations in goal margin prediction prevented real-world profitability. The model was tested by simulating betting as monte-carlo-simulations and visualizing it's accuracy compared to simpler betting paradigms.
DecaDruck (Practice project)
Categories:
Product
Frontend
Backend
Branding
Online Shop
Real time Calculation
Logo Design
[Practice project] DecaDruck aims to open 3d printing to people without technical knowledge and industry expertise by combining consultation, design and realization into one service. If vistors already have a 3D-file, they can calculate the printing cost, with volume discount included.
Download Resume (German)
