My name is Niklas, and I am currently based out of Berlin where I work as a data scientist in the logistics department at Delivery Hero. In my role I mainly deal with predictive modeling around the topic of optimal courier staffing throughout the day. Before moving to Berlin, I studied electrical engineering at RWTH Aachen University where I graduated with an M.Sc. in 2014, and a doctoral degree in 2019.
Pymanopt is a Python toolbox for numerical optimization on Riemannian manifolds with support for automatic differentiation. Riemannian structures frequently arise in the context of signal processing and machine learning where constraint sets often form a smooth nonlinear search space. To attack such problems, Pymanopt, which is based on the MATLAB toolbox Manopt, provides a variety of different solvers (e.g. steepest descent, conjugate gradients, trust-regions), manifolds (e.g. Stiefel manifold, Grassmannian, positive definite matrices) and autodiff backends (currently Autograd, TensorFlow, PyTorch and JAX). By appealing to automatic differentiation, the toolbox lowers the entry barrier for rapid prototyping as the need for laborious calculation of higher-order derivatives of matrix expressions by hand is removed.
Lampions is an email relaying service leveraging AWS SES to forward incoming emails on a verified domain to pre-configured forwarding addresses. The goal is to insulate a user’s primary email address(es) from third-party services by providing an easy and convenient way to create per-service forwarding addresses. The project consists of a command-line utility to configure the necessary AWS infrastructure and define email routes. Additionally, a browser extension can be used to define, remove or temporarily deactivate email aliases in case one wants to stop receiving emails for certain aliases.
During my teenage years, my creative outlet was editing frag movies about the competitive scene of the game Counter-Strike. I eventually retired the hobby in 2008 after enrolling in university but picked it up again for a while in late 2015. Some selected works dating back all the way to 2007 can be found below.