Transforming the world with curious minds

We conceptualize, build, and scale AI ventures.

Merantix is a research lab and venture builder in the space of artificial intelligence. We help mid- to large-sized companies to explore and leverage the potential of deep learning. We have a strong bias for partnerships with companies that have powerful data sets, and relatively short decision cycles. Consulting fees are not our main incentive, in contrast we are willing to take investment risks together with our venture partners.

We build partnerships with leading companies and organizations to assemble complex datasets and build machine learning solutions that address industries such as health, finance, automotive and advertising.

Icones How #3 Merantix (SHARED) 170228

We iterate to transform models into … [more]


We iterate to transform models into standalone products. This means scaling and generalizing models, building stable APIs that support client needs, doubling down to ensure data security, and more!

Icones How #2 Merantix (SHARED) 170228

We build partnerships with leading companies … [more]


We build partnerships with leading companies and organizations to assemble complex datasets. We target data science challenges that go beyond rule based solutions or human capabilities.

Icones How #4 Merantix (SHARED) 170228
Machine Learning

We develop models that predict actionable … [more]

Machine Learning

We develop models that predict actionable insights. We use state-of-the-art deep learning techniques and infrastructure and apply rigorous measures of success.


When products reach a certain maturity … [more]


When products reach a certain maturity and critical mass of users, we spin off companies with dedicated teams.


We form joint venture partnerships in which we serve as an artificial intelligence laboratory and venture builder. We leverage our expertise building machine learning infrastructure and applications and combine it with our partners’ datasets, their domain specific knowledge and market access. We apply cutting edge research to real-world problems, conceptualize, build, and scale ventures that deliver unmatched value in many of industries.



The financial services industry has a long history of taking advantage of new technologies, and deep learning is no exception. Today’s financial institutions use artificial intelligence to improve prediction models, analyze cash demands and trading. Merantix provides cutting-edge technology to support our venture partners in their effort to stay ahead of the curve.

High Frequency Trading

Merantix built an AI-based system capable of predicting currency exchange rates at market relevant speeds. Our deep models recognize complex market movements way ahead of traditional methods. Through our partnership with an experienced HFT trading firm, we are able to deploy the models on the latest infrastructure making sure that we can compete at the forefront of high frequency trading. We see this as a great playground to test out state-of-the-art research of deep learning as for HFT it is all about squeezing the last bit of performance out of the models.

Credit Scoring

In partnership with Intrum, one of Europe’s leading credit and collection agencies, Merantix developed technology to significantly improve the distinction between good and bad payers. In comparison to current scoring methods, our deep learning algorithm was able to reduce the amount of rejected purchases by 30%, while the general risk of default remained the same. This may ultimately not only reduce the default rate and thus cost but may also increase the number of positive decisions and thereby lead to additional revenue.

Key figures

30 %
We were able to reduce the amount of rejected purchases by 30%, while the general risk of default remained the same.
Data points considered per day
2 M
Our model evaluates more than 2 M times a day to find the optimal trading moment.


As more data about genetic preconditions, biological biomarkers and lifestyle factors are created and linked to electronic health records, a new era of healthcare is on the horizon. Deep learning will be a driving technology pioneering new models in health and life sciences. Moreover, it increases the accuracy and quality of diagnosis and clinical treatments in various areas and has the potential to drive precision medicine as it is capable of dealing with vast amounts of data gathered from mobile devices, genomics and medical health records.

Medical Imaging

Merantix is building an AI-backed service that performs automatic diagnoses of radiology images. To train our deep learning system, we extract gold-standard labels from given natural language text reports. This is an ongoing task and is being repeated for different use cases (X-ray, CT, MRT, etc.) and can be applied for imagery of varying body parts. We further automate and streamline the image labelling process, to improve efficiency and allow doctors to spend more time with their patients.

Alzheimer's Early Screening

What if you could detect Alzheimer’s up to 6 years prior to its onset, using an iPhone only?

We helped ETH spin-off company Altoida to develop and productize a methodology that is based on longitudinal clinical studies with more than 2700 Alzheimer’s Disease (AD) patients over 8 years. After taking an iPhone app based test for 10 minutes, Altoida is able to to tell you if you will develop AD within 6 years from now with an accuracy of 94%.

Patient Data

Merantix provides an efficient and automated approach to the analysis of large amounts of patient data. Using blood, urine samples and other accessible medical records our machine learning engine is able to support physicians in the detection, treatment and surveillance of a variety of conditions.

Key figures

94 %
Altoida technology can diagnose Alzheimer’s disease up to 6 years prior to onset with 94% accuracy.
Medical Lab tests
200 M
We work with a data set of more than 200 M medical lab tests.
98 %
We can automatically structure medical reports with more than 98% accuracy.


There have been massive improvements in autonomous driving, and other related capabilities in the automotive industry in recent years. Driven by the pioneers in the industry, full market adoption of these technologies is approaching fast. Deep learning is a major enabler in the space of autonomous driving systems and is used for various tasks including the detection of other vehicles on roads, classification of traffic lights, road signs and obstacle recognition. However, in order to make machine drivers safer than human drivers there is still some way to go. Experts agree nonetheless that deep learning will take this hurdle very soon.

Feature Visualization

In partnership with Bosch, Merantix is working on evaluating and improving deep learning tools in autonomous driving. We are creating, training and evaluating neural nets that recognize objects, such as traffic lights, signs and cars, from driving assistant system sensor data (i.e. videos and images). Our system visualizes the characteristics of a trained neural network based on state of the art research. Besides visualizing the parametrization of the neural network itself ,the tool is also able to highlight relevant regions in an image which led to decisions by the neural network and quantify those.

Key figures

Number of papers
published in top-tier journals and conferences on visualizing neural networks.
Exposed software
of automakers do not have a hacking countermeasure strategy in place.
Growth rate
faster growth rate of the connected car market vs. the total car market.


Recently deep learning has entered the creative space and sparked surprise by painting in style of impressionist artists or by generating music. Until now, technology was often used to support humans in their artistry, soon however we will see more and more songs, paintings and performances entirely created by machines. Online advertising provides an ideal environment for AI-based systems to enter the creative space, as the success of the creative work can be quantified by traction.

Creation Automation

The long term vision of this Merantix lab project is to fully automate the creative generation while improving conversion. Our model, developed in collaboration with Rocket Internet, analyzes the creatives designed before they go live on Facebook and predicts their performance in terms of traffic to detect recurring patterns. In the long term, we envision that this model helps us to automatically generate creatives.

Key figures

Impact of creative on performance
Estimations say that around 30% of the performance of an advertisement depends on the creative.
Number of creatives
1 M
We train our models on more than 1 M creatives to really understand what makes an creative stand out.

Adrian Locher

Adrian is a technology serial entrepreneur and founder of Merantix. He built and scaled his last business to a leading e-commerce player with more than 100M revenue and 200 employees. Adrian graduated from University of St. Gallen (HSG) in Business and Economics.

Rasmus Rothe

Rasmus is one of Europe’s leading experts in deep learning and founder of Merantix. He studied at ETH Zurich, Oxford, and Princeton. Previously he launched as part of his PhD research, started Europe’s largest hackathon HackZurich, and worked for Google and BCG.

Josh Chen

Josh is a Machine Intelligence Engineer at Merantix. He graduated from Princeton with summa cum laude. Prior to joining Merantix, he worked in machine learning positions at D.E. Shaw & Company and Amazon.

John McSpedon

John is a Machine Intelligence Engineer at Merantix. Prior to working for Facebook and Sift Science, he graduated in Computer Science from Princeton.

Ryan Henderson

Ryan is a Machine Intelligence Engineer at Merantix and holds a PhD in Chemistry from Cornell University. Prior to joining Merantix, he worked for Intel and a Berlin based Blockchain company.

Luc van Gool

Prof Luc van Gool is an advisor to Merantix. He leads the Computer Vision Lab at ETH Zurich in Switzerland and the Institute VISICS at the University of Leuven in Belgium. He has been the advisor to more than 10 computer vision startups and serves on the board of various top tier journals and conferences.

Radu Timofte

Radu Timofte is an advisor to Merantix. He is a Research Group Leader at the Computer Vision Lab at ETH Zurich in Switzerland. He has published many high-impact papers on Computer Vision and (Deep) Machine Learning in top-tier journals and conferences.

Arne Bleckwenn

Arne is a Partner at Merantix. He is a serial entrepreneur and angel investor. After building and exiting an online gaming company, he has built AirBnB competitor Wimdu and raised 90M EUR venture capital. He graduated from WHU with an MBA prior to joining McKinsey.

Jonas Muff

Jonas is Head of Business Development at Merantix. He graduated with a Bachelor’s degree from University of St. Gallen (HSG). During his studies he was CEO of START Summit, a student-led entrepreneurship conference with over 1000 participants. He is also the initiator of START Fund (part of HSG Foundation).

Combining your industry expertise with our machine learning skills,

for a successful co-creation of new business opportunities.

Facts not bullshit

100 M €

Annual revenue of last venture

200 +

Peer-reviewed research papers

150 M

Images uploads on

2000 +

Hackers at our Hackathons

Your data and knowledge,

Schedule an appointment

Meet us
Merantix GmbH
Friedrichstraße 68
10117 Berlin, Germany
Registration No.
HRB 176845 B
Managing Director
Dr. Rasmus Rothe

We are growing,

we are looking for the most talented curious minds out there.


Machine Intelligence Engineer

Berlin, Germany