deep learning in the wild
Deep Learning in the Wild

Luminovo.aiBespoke AI solutions

Who are you and what’s your role in the company?

Timon: My name is Timon Ruban and I am the Co-Founder and CTO of

Sebastian: I am Sebastian Schaal, the second Co-Founder and CEO of

How did you get in touch with deep or machine learning in general?

Timon: I first studied electrical engineering at ETH Zurich and had actually been pretty oblivious to the whole machine learning hype before I arrived at Stanford for my master’s degree. There I took Andrew Ng’s ML class and quickly realized that all the math I learned at ETH could actually be put to good use to solve real-world problems using machine learning.

Sebastian: My journey towards ML was in some way similar to the one of Timon. I went out to Stanford to further position myself at the intersection of Business and Tech. However, as most people there it did not take long until I got dragged into the force field of ML. During the second half of my program I tried to take all ML classes Stanford had to offer and when I returned to Munich to finish my second master’s program, I made sure to follow the classes I missed during my time in California (like Karpathy’s CS231n-class on CNNs).

Describe what your project/business does. Which problem are you solving?

Sebastian: When we came back from the Silicon Valley, we found that the adoption of AI is still lacking in many German companies. We think this is due to the fact that 1) AI often requires a new way of thinking about problems and 2) that there are only few people with actual experience in the field able to push this adoption forward. Since Timon and me wanted to apply deep learning in the wild anyways, we decided to start to help companies kickstart their AI efforts.

At, we often start with helping companies get their thinking straight about which of their problems can be tackled with machine learning and the data they have. Then we deliver functional AI prototypes that can be directly deployed and hopefully generate value for our clients immediately. In contrast to others, our focus is on enabling our client’s employees to get started with AI on their own after we are gone, so instead of selling black boxes and licensing fees, we share our code, documentation and best practices and do workshops at the end of every project explaining our methods and ways of approaching their problem.

Can you describe your tech stack, what libraries hardware / architecture are you using and why?

Timon: We are all Pythonistas at Luminovo, so we mostly use TensorFlow (I got addicted to TF Slim during my time at Google) and train our models on Google Cloud. For the occasional traditional ML problem, we usually fall back to pandas/scikit-learn.

What were some of the biggest challenges technical or business related you encountered and how did you overcome them?

Sebastian: I think, as a young startup, getting the initial credibility is always tough. It is the classic chicken-and-egg-problem: you need references to close large clients, but you do not get good references without large clients. In this situation, having a decent network and, even more important, people who promote you, is invaluable. Most of our initial customers came from warm introductions, bringing us directly into a meeting with the stakeholders. Another thing that is always a bit of a hassle in Germany is getting the data privacy setup right. There are just a lot of things you have to consider when you work with sensitive data.

How has the industry response been to your solution?

Sebastian: We have been thinking a lot about our offering and what would be appealing to us and our clients. I think our USP at the moment is also something our clients value a lot: we are not selling licensing fees or locking in our clients with expensive support contracts. When we take on a project, our goal always is to transfer the knowledge to the employees of our clients.

What excites you most about the future of deep learning? Where do you hope the technology is going / where do you see the greatest opportunities?

Timon: I think going even more meta than deep learning already is, with approaches like Neural Architecture Search, is a really interesting trend, but it seems like that will only be feasible for the non-Googles of this world (like once the hardware gets even faster (Graphcore? Intel? Show me what you got!).

How do you keep up with new papers and publications?

Timon: I read newsletters like Deep Learning Weekly and Import AI and do a daily check on r/MachineLearning. In addition, I always just hope my friends are kind enough to let me know when they read a really cool new paper.

What do you wish you had known before getting into deep learning?

Sebastian: Before getting into the field, I was not aware of the variety of problems that could be tackled with neural networks. Also, I feel that a lot of people are hesitating to get started since they think they are just too far behind. However, you can already have some meaningful impact with just having domain expertise and a sound understanding of how to tackle problems with DL.

What do you wish to see on a candidate’s resume who is applying for an engineering position involving deep learning?

Timon: To me the scarcest resource in deep learning today is still experience. So if an applicant can highlight with his resume that he already used machine learning to solve real-world problems (this can even be within the scope of a really interesting class project), that is always a very good sign. In terms of online courses, I’m a big fan of CS231n and’s new 5-part deep learning course.

What do you recommend candidates do to make themselves more attractive to employers looking to hire a deep learning professional?

Timon: Pick a problem/dataset you think is interesting and try to re-implement a relatively “state-of-the-art” model, put your code/results on GitHub and maybe write a Medium post about what you learned. That should take a few weekends at most and will give you a great conversation starter for any interview for a deep learning position. Other than that, reading papers and being able to discuss the most recent neural net architectures/trends always makes you look like you know what you are doing.

Can you give us some facts about your current team and point out, why a candidate may want to join you?

Sebastian: Besides Timon and I, we had Sebastian Schuon on board from the very beginning as our chairman. Through his successful exit of Stylight to ProsiebenSat1 and his role as an active angel investor after that, he brings quite some entrepreneurial experience to the team. In February, Oskar Triebe, a friend we still know from Stanford, joined us as the Head of Business Development. On the tech side, we are currently working together with some highly talented students from the TU Munich.

For us, working at is an almost unique learning opportunity. Just like us, you will have the chance to deploy AI in the wild in a variety of projects across different industries. This will give you a better understanding of the diversity of challenges out there than working within one company on similar problems and datasets. We think that developing an intuition will even be more valuable in the future, when using and training models will become more and more of a commodity.