With planovia.com we tackled the problem that architects and other people cannot get a Schwarzplan (a map where just buildings are mapped) for free to download. So we simply used given APIs and computing power to make it available for everyone to get his personal Schwarzplan from all over the world.
During a short vacation stop in Wesenberg (a city in germany, on canoe trip) a friend and I had some deep discussions on a grill that fits in every bag. Our idea and the USP of the product was the packing size and the design of the product. So we iterated and reconsidered the product in every step. We’re proud to present the A4Grill. The grill is also known as Mehrweggrill. Since 2015. Agentur B&F is a Dresden based agency that is specialized on producing and commercialization of different products. Beside the A4Grill the agency also provides a variety of platforms and blogs as well as consulting services. Since 2016.
NLP and machine learning was/is one of my passions. Under the flag of trendinc I train, apply and use statistical models. Some applications are news aggregation and processing (sentiment, author level, media buzz etc), trend discovery, conversational bots and a question answering system that I’m developing since 2015. With trendinc I’ve created a label to ship prototyped apps and create fancy software products.
With Hack.Events we established hackathons and fancy startup events in the area of Dresden. We are currently planning more of it, to enhance the startup scene in Saxony. Since 2016.
# Technical Book Remediation
For the mitp-Verlag one of the biggest tech publishers in Germany, I did the technical book remediation for two Deep Learning books. First was from Francois Chollet about Deep Learning in Python using Keras framework (Google Books). The second book was the (so called) standard reference about Deep Learning from Ian Goodfellow, Yoshua Bengio and Aaron Courville (Google Books coming).
# DecompileD Conference
Starting in 2018 we created DecompileD Conference to have a first class international conference on hot topics in tech. We started with an intersection of: machine learning, mobile and cloud. I’ve accompanied this process as lead organizer. Since 2018.
# Startup Weekend Dresden
Together with a few partners like GründerGarten Dresden, Stadt.Land.Netz and ImpactLoft we (Hack.Events) organized the 2017 version of the Startup Weekend in Dresden. The event took place in December. Since 2014.
# MLDD – Machine Learning Meetup Dresden
With #MLDD Meetup we established a spot for machine learners in saxony (esp. Dresden). The Meetup talks place nearly every month with presentations ranging from theoretical data science to practical machine learning applications. Beside the talks we often have poster sessions and interesting discussions.
# Emmy – Autonomous RC
We (a friend and me) are working on a framework for an autonomous RC car. For this we are limiting ourself to just one camera. Currently we are trying various approaches, from very shallow nets up to sophisticated deep learning stuff.
Is the location of a city or location in germany predictable? DreamCity deals with that question and provides a regressor that estimates the lat, lng coords of different german cities. This former research project was released as open source project to play around with it.
# Gopher Neural
When I started with machine learning in Golang, I searched the web for a framework that covers all aspects of ML: training, evaluation, production. I found go-neural – a simple back propagation algorithm. So I built everything I needed around this algorithm. The product was Gopher Neural. Check it out on Github.
Most phonetic algorithms using a code or symbols that are not easily comparable. This motivates the development of a phonetic algorithm that uses the simplest forms of text comparison. With Phonetizz the Levenshtein distance metric was utilized to perform such calculations for such a phonetic measure. Another important point is that all known algorithms are just working on words. With Phonetizz it is possible to work also on phrases instead of just one word.
# Question Answering using Unstructured Data
During my master thesis I was working on this topic. The task was to build a pipeline that brings unstructured text into a question answering system. I’ve tackled this topic by using a common pipeline and enriched this pipeline with on trained classifiers. After finding all possible answer candidates, the system also used machine learning models to decide für the right version after scoring. Find article, thesis and repo here.
# whisp – The opinion network
Whisp structures opinions from users and recommends relevant stuff to you. The range of topics is wide. Movies, music and apps are just a few. Whisp was developed for iOS and Android. Also a web version is available. From 2011 to 2014.
# Ipsum – PHP formular parser
The parser written in PHP parses a formula given as string and calculates the result. You can also add own functions, variables and constants to the calculation.