Henning Sperr
Full Stack Data Scientist
View My LinkedIn Profile
Portfolio
Tech stack
Work Projects
Senior Research Engineer at Zalando (~15.000 employees), Berlin, Germany Since Jun 2018
Fraud detection and prevention
- Technical lead of a team of 7 people; align and structure team’s analysis and documentation process to increase team’s reproducibility and efficiency.
- Planned, developed and implemented a novel fraud detection system replacing the previous manual screening at a higher precision and recall.
- Defined and owned Objective Key Results which had a major impact on company direction in terms of machine learning. Setup of a new reporting structure to product management and C-Level Executives.
Data Scientist/Head of Data at PAIDY (~20 employees), Tokyo, Japan Aug 2015 - Mar 2018
Online payment credit scoring and fraud prevention
- Founding member and initial lead of the Risk team. Responsible for credit underwriting and fraud prevention models. Main achievements:
- Reduced credit default by 66% saving the company millions of dollars.
- Reduced response time to emerging risk trends to hours, down from multiple days/weeks.
- Designed, deployed and maintained a credit risk monitoring system that reduced the response time to emerging risk trends from multiple days/weeks to hours.
- Responsible for full stack, from planning, selecting features, to building and deploying models, credit policies and risk related processes.
Data Scientist at Rakuten (~10.000 employees), Tokyo, Japan Oct 2013 - Aug 2015
Product Targeting and Recommendation
- Won Rakuten award for significantly improving advertisement revenue
- Employed unsupervised machine learning algorithms to cluster similar items and perform ad-keyword selection, and implemented system to allow safe data access by other company sections.
Research Assistant at Karlsruhe Institute of Technology, Germany Aug 2011 - Jun 2013
Working in Cognitive Sciences and Semantic Knowledge Management
- Developed the semantic knowledge management tool iMapping (commercially available)
- Worked on Japanese-English machine translation system and various speech recognition tasks.
Projects
Halite III - Deep Reinforcement Learning
A series on how to implement deep reinforcement learning on non openAI gym examples. We have to write everything from scratch and come up with our own rewards and state representations.
We implement a simple game from scratch and implement minmax/MC tree search and eventually want to implement AlphaGo style algorithm to play the game.
While in Univeristy a friend challenged me to write a chess engine in C. So here it is, while the code is not the cleanest its a fully functioning chess engine that will beat the average chess player.
A wrapper around python datetime to have convenience methods similar to javas Joda time. For what I needed it, it works but I would not recommend using it. I should have wrapped this around pyarrow or in general just not touch datetime libraries. I know it has some bugs depending on your usecase.
A simple Anki plugin which will insert the english meanings and katakana spelling for kanji inserted into Anki. This makes it a lot easier to insert many words at once.
I don’t kaggle much, but one competition I entered and participated in the community I contributed a kernel that helped many people ensembling models and netted me being featured on the kaggle blog.
Page template forked from evanca