SOTA multi-modal product matching with foundation models
SOTA multi-modal product matching with foundation models

We present an end-to-end system that can identify different representations of the same garment across outfits, models, backgrounds using multi-modal deep learning and a human-in-the-loop pipeline.

AI computer vision foundation models

Human and Algorithmic Pricing
Human and Algorithmic Pricing

Using data from a large field experiment, we show that algorithms price more efficiently than humans, but a system that allows effective collaboration between humans and algorithms prices best of all.

economics AI pricing

Interference in Pricing Experiments
Interference in Pricing Experiments

Many companies rely on A/B tests to experiment on different pricing policies. In this working paper we argue that customer substitution behavior leads to substantial bias, inflating false-positive rates of experiments, and we explore remedies.

causal inference experimentation pricing

Data Science Soft Skills
Data Science Soft Skills

This blogpost summarizes what I've learned to solve real-life problems end-to-end, from business requirements to a convincing presentation of results

data science leadership culture

Claims data in healthcare: An overview
Claims data in healthcare: An overview

This legal journal article gives an overview of healthcare claims (reimbursement) data and its potential for social good.

advisory healthcare

Causal Inference in Industry
Causal Inference in Industry

This article explains an end-to-end application of machine-learning-powered causal inference in industry, building on my experience in pricing.

causality machine learning analytics

Technology against COVID-19
Technology against COVID-19

As an invited expert, I co-authored a discussion paper by Germany's National Academy of Sciences Leopoldina. The group of authors unites an impressive interdisciplinary expertise, allowing for a holistic view of digital opportunities in COVID-19-response. Particularly exciting for me personally: all available evidence suggests a positive, if moderate, effect of Germany’s Corona-Warn-App.

COVID-19

For more policy stints
For more policy stints

National business newspaper Handelsblatt wrote a profile article on my policy stint, where I explain what motivated me to spend time in government.

press agile government

Querwechsler Netzwerk
Querwechsler Netzwerk

I became a member of policy advocacy group "Querwechsler Netzwerk", a group of experienced managers who believe that public sector stands to benefit from industry hires.

advocacy

The Story of the Corona-Warn-App
The Story of the Corona-Warn-App

Germany was among the first countries worldwide to launch a privacy-preserving COVID-19 exposure notification app, and I had the privilege to be part of the team. In this talk, I explain some key decisions; and I ask what governments can learn from the experience.

COVID-19 health data agile government

Vorsprung Durch Devolution
Vorsprung Durch Devolution

Despite a remarkably successful Covid-response in Germany, the grass is always greener on the other side. In this opinion piece, Axel Heitmueller and I discuss healthcare decentralization envy, to contrast pandemic readiness in Germany with the UK.

policy healthcare covid-19

Hacker's Guide to Healthcare
Hacker's Guide to Healthcare

Speaking at Europe's largest hacker conference, 36C3, I provided an overview of ways for the hacker community to get involved in healthcare. This was a very exciting experience, connecting my work in policy with my dream of speaking at a chaos communication congress.

data science healthcare

AI: Cherry on the Cake
AI: Cherry on the Cake

This summary article offers my view of and prediction for AI in healthcare: less grandiose hype, more real quality of care. It was published in the magazine of Germany's largest payer organization.

artificial intelligence hype

Hype-based medicine
Hype-based medicine

Switching from hands-on data science into policy advisory, I was surprised about the amount of hype surrounding artificial intelligence and blockchain. In my first opinion piece, I argue for more attention to the "boring" task of collecting quality training and testing data.

artificial intelligence blockchain

Probabilistic Forecasting
Probabilistic Forecasting

This project used state-of-the-art forecasting ideas to provide probabilistic, as well as counterfactual, forecasts of retail demand across thousands of products.

pytorch lightgbm RNNs time series

Deep Causal Inference
Deep Causal Inference

Parsing causal signal from observational data is notoriously difficult. This project attempts to get good estimates for price elasticity of demand in a setting typical for ecommerce and retail.

tensorflow causal inference

Humanitarian Mapping with Deep Learning
Humanitarian Mapping with Deep Learning

In situations of humanitarian disaster, available map material is often not good enough for efficient relief efforts. This project explored convolutional neural networks to pre-classify images for the humanitarian openstreetmap team.

python CNNs caffe

Angrier Birds: Bayesian Reinforcement Learning
Angrier Birds: Bayesian Reinforcement Learning

We train a reinforcement learner to play the game Angry Birds. We show that the efficiency of regular ε-greedy Q-Learning can be improved through more systematic exploration in Randomized Least Squares Value Iteration (RLSVI), an algorithm that samples its policy from a posterior distribution on optimal policies.

python reinforcement learning

Constitutional Precedent of Amicus Briefs

We investigate language in U.S. Supreme Court majority opinions that originate from interest groups' amicus briefs. For this work, we compiled and analyzed a large corpus of legal text to track the origination of concepts and ideas gaining precedence.

python webscraping legal tech

Sentiment as a Predictor of Wikipedia Editor Activity
Sentiment as a Predictor of Wikipedia Editor Activity

We perform sentiment analysis on messages exchanged between Wikipedia editors in the so-called "user talk pages," to predict future user editing behavior. Our findings could be used better engage editors, potentially resulting in better article quality.

NLP python

And1: Skip to My Loop!
And1: Skip to My Loop!

We built a "basketball"-playing robot, competing against 35 other teams after just under three weeks of construction.

robotics Arduino C

Evolutionary Extortion and Mischief: Zero Determinant Strategies in Iterated 2x2 Games
Evolutionary Extortion and Mischief: Zero Determinant Strategies in Iterated 2x2 Games

My BSc thesis in economics explored a surprising discovery in iterated 2x2 games. I showed that extortion is feasible in more games than just Prisoner's Dilemma, and simulated convergence behavior. I am proud that as an early and accessible summary, my paper was cited by several researchers.

game theory C++

Insurance for Justice: Social Welfare under Uncertainty
Insurance for Justice: Social Welfare under Uncertainty

My BA thesis in philosophy sketches a justification of a welfare state, defending it against both givers and receivers of redistribution. To finish my studies in philosophy, I undertook this ambitious project to argue for concrete and relevant results. The thesis is written in German.

Dworkin luck egalitarianism

Isometric strategy game engine, 2004
Isometric strategy game engine, 2004

My interest in coding started early: In high school, I did a lot of coding for fun. Perhaps my most involved project was building a 2D-game engine from scratch. I'm still proud that I managed to make little animated figures find their own path to walk from A to B to build little skyscrapers on grass.

Code