Jonathan Masci: Team building and communication

Jonathan Masci – Managing Director of Quantenstein GmbH

Quantenstein was one start-up participating in the “Inception Connect Enterprise” event from NVIDIA. With the “Inception Connect Enterprise” event, NVIDIA creates a platform for AI startups, investors, industry leaders, and others from the Tech community to demonstrate the value of artificial intelligence across all business sectors.

With the Inception program, NVIDIA supports AI start-ups on various levels to help them during the critical phases of product and prototype development and deployment. This is done, among other things, by providing technical support from deep learning experts, hardware discounts and marketing.

Please introduce yourself and the Startup Quantenstein to our readers!
Jonathan Masci: I am General Manager at Quantenstein GmbH, and co-founder of NNAISENSE SA, the AI startup that is developing the Quantenstein’s underlying technology. My background is in machine learning, AI, and computer science, and I have always been fascinated by finance. I received my PhD in Machine Learning and Deep Learning at the University of Lugano and the Swiss AI Lab, IDSIA, under the supervision of Prof. Juergen Schimdhuber.IDSIA has been at the forefront of many of the breakthroughs that have ignited the currentboom in AI, and has recently been awarded the prestigious NVIDIA Pioneers of research award. TheNNAISENSE founders all met while working at the institute, and the mission of our company is to scale and accelerate the success of our former lab.

Quantenstein is a joint venture between NNAISENSE and ACATIS Investments,based in Frankfurt thatprovides marketingand finance expertise based on their20 yearsof success in the financial sector.The focus of the company is Long-Term Value Investing (LTVI) for which we have developed, over the course of almost two years, cutting edge AI and Deep Learning technology. The system is fed the time series of company fundamentals (the same information commonly used by discretionary portfolio managers) and is trained to produce optimized portfolios that maximize the Sharpe or Information ratio, and that are fully customizable in terms of sectors, regions, volatility constraints, etc.

Unlike other financial managementtools, which are usually designed to aid the decision-making process, our system has no human in the loop, directly taking decisions on its own,and is therefore less biased (e.g. no “gut feeling”).

How difficult was the start and what challenges you had to overcome?
Jonathan Masci:Nothing is easy at the start, and I would have been surprised if it had been a journey without challenges, this would have in fact meant that the financial market is highly predictable, and we all know it is not.
It is never a good idea to immediately shoot for the final product, but instead have incremental milestones.Too many things can go wrong and trying to fixing them would be like finding a needle in a haystack. That’s why we started with stock picking, the problem of finding companies that are good at some pre-defined task (e.g. outperform the MSCI World,high returns, etc.).
The problem is that Deep Learning requires very large quantities of annotated data, which, unfortunately, is not available when doing LTVI. Also,you only get to see one history, and there is not an easy way to simulate analternative past. But once you have a hammer—DL in this case—everything looks like a nail, so we spent considerable time and effort in trying to cope with data shortage, and making the system as robust as possible, which eventuallyledto improved performance.
It turns out, however,that we were hammering the wrong nail because building portfolios in an end-to-end fashion, rather than just doing stock picking,gives us plenty of more data to work with. You can select a random subset of the available companies at any time and this gives a combinatorial increase in the number of possible training trajectories. It also requires undertaking the major research task of developing new algorithms to be able to process this type of data since current deep learning approachesare not designed for it. This is related to the application of deep learning to graph and set-structured data, one of my active research areas at NNAISENSE.

Who is your target audience?
We target financial clients and we have just started a global equity fund with BayernInvest called “Acatis Ki Actien Global Fonds”.

What is the USP of your startup?
Our unique selling point is the fully-automated end-to-end portfolio construction that can be customized to optimize the customers desired investment criteria. Stock selection and portfolio construction are usually two separate components in any quant strategy, we instead learn them jointly.

Can you describe a typical workday of you?
I usually try to wake up reasonably early, even if I usually work until late when I need to do coding or think on how to solve the next problem. After a quick morning meeting with the team I respond to all emails (I try to have a zero inbox policy), check out any relevant new scientific paper and then move to the tasks which I have for the day. That can span coding new features, designing new deep learning models and making sure that everything progresses as close as possible to the plan.

Where do you see yourself and the startup Quantenstein in five years?
At NNAISENSE we are already expanding the range of financial applications to increase our offering. Short-term trading will be one of those, we plan to have a test ride before the summer. In five years, I see myselfstill actively contributing to set directions for the core research team, and more active at promoting the company.

What 3 tips would you give other Start-up founders on the way?
1. Get things done rather than endlessly agonizing over what is the optimal strategy from now until the end of time, keep an open mind and be ready to reconsider your decisions.
2. Team building and communication. At the early startup stage, it is still possible to know about everything that goes on, and this amplifies the rate at which people collaborate and get things done.
3. Research, invest resources to always be ahead in your field, the fun starts when the minimum project requirements are achieved.


More information you will find here

Thank you Jonathan Masci for the Interview

Statements of the author and the interviewee do not necessarily represent the editors and the publisher opinion again.

Sabine Elsässer

Sabine Elsässer ist 39 Jahre jung, Gründerin und leitende Redakteurin der StartupValleyNews. Ihre Karriere startete sie in verschiedenen internationalen Direktvertriebsunternehmen. Seit 2007 ist sie hauptberuflich als Journalistin tätig. Während dieser Zeit lernte sie die Startup-Szene kennen und schätzen, was Sie dazu bewogen hat mit StartupValleyNews ein internationales Startup Magazin aufzubauen!


Sign up now for free and be the first to get the newest Startup news