-
At Hopsworks, we always aim to push the state of the art forward✌️
Moritz Meister
Software Engineer

Your background:
Contrary to most people at the company, I didn’t start my career with a traditional Computer Science degree. Instead, I did my Bachelor degree in Econometrics and Operations Research which is more similar to an applied mathematics/statistics degree. However, we were also taught Computer Science basics like Data Structures and Algorithms.
After working two years as a Data Scientist for analytical customer relationship management, I knew I wanted to move from the modelling side to the tooling. Thus I decided to do my master degree in Computer Science at EIT Digital Master School. The main focus of the degree was scalable distributed systems for data processing, and I think this was a great decision because building tools for Data Scientists is much easier and a lot of fun if you have worked as a Data Scientist yourself before.
What's your current role?
I am currently leading the Hopsworks Feature Store development. That means I spend most of my time working on the backend of the Feature Store and the HSFS Client SDKs. Apart from that, I also lead efforts to structure the work of the team when it comes to the planning and design of new functionality for the platform.
How did you join Hopsworks?
During my Master degree, I came to Stockholm for the Big Data Summer School at KTH University and our CEO Jim Dowling, who is also a professor at KTH, gave a talk about the very early version of Hopsworks. I liked his vision for the platform, so I applied to do my master's thesis internship at the company.
After the successful completion of the thesis, I was offered a job as a Software Engineer.

What project are you currently working on?
The online feature store of Hopsworks keeps only the latest feature value for each feature and entity. However, there are use cases where data needs to be purged from online storage in order to limit storage requirements but also because data might become stale and therefore shouldn’t be used for predictions. To accomplish this I am currently working on Time-To-Live policies for the online feature store.
Favourite programming language?
Python, because it’s the most flexible language to implement useful and abstract User APIs.
In the long run, are feature stores part of broader ML platforms or standalone product?
I think feature stores will be part of broader ML platforms. A feature store goes hand in hand with model serving, which again goes hand in hand with model training. Since it is cumbersome to write code to bridge all gaps between the different stages of the end-to-end ML pipeline, I believe the community will converge to a few broader platforms that will cover most of the use cases. There will always be unique cases, of course, which will require custom tooling.
Why should someone want to join your team?
At Hopsworks, you have the opportunity to take ownership of a part of the product that you are passionate about. The research background is still a core part of our culture, therefore effort is always credited and valued. Furthermore, we always aim to push the state of the art forward.