Interview with Maarten Masschelein
CEO, Soda - A tool for data observability.
What is Soda? Why did you build it?
Soda is a data observability company that provides open source data reliability engineering tools and a data management platform.
In our day-to-day lives, we are all using data more than we realize to make decisions and our choices are driven by data, and in business, almost all of our decisions - even the micro-decisions - are all heavily influenced or enabled by data.
As we’re seeing this shift, data as a practice is still not as mature as what we see with software engineering. Data is lagging behind about a decade or so, notably in practices such as testing and monitoring. Without proper testing, as new data comes in, things can fundamentally change with that dataset, think structure, meaning, and completeness, leading to data issues with the reports we use or the machine learning models we train.
Sometimes the dataset may not be refreshing, which is the worst problem. You might not notice and your reports or data products will stop working, or they will continue working, but with bad data. And that’s why we know that the biggest problem faced by data teams today, is that they are flying blind. Without a clear strategy to monitor data for issues, most organizations do not know how to start addressing the problem which leaves their systems exposed and can result in serious downstream issues for the data products they are building.
We have been in those situations where you build data products for teams that need to do a lot of analysis at the end of a month or quarter. They are typically in functions where they need to report KPIs at the end of a business cycle, and all of a sudden they notice that the report is broken and they cannot use it. That’s a very frustrating feeling. It’s really all about the ability to trust the data that you have in your organization.
That’s the core of what we’re solving for at Soda: giving data teams the power to create trusted data products, and an integrated platform to bring everyone closer to data.
How have engineers solved this in the past?
There have been some tooling solutions in the past. If you go back in history, some of them got acquired by the likes of Oracle, Informatica and IBM. They were tools that were tightly coupled with the data transformation layer and tended to work as rule-based systems.
For example, you could specify certain rules like value ranges for each column which would then trigger alerts if the data differed. This often involved a very technical interface that was tedious to use, and you had to know all of the business logic up-front. There was no automation.
Another way of solving this was to manually sift through samples of data at regular intervals, such as every Friday or at the end of the month. You would just take a dataset, grab a couple of records and sift through it to see if there were any issues that could be identified.
Scheduled queries are another stop-gap solution we often see, where dashboards are built for a set of regular queries. But again, they are manual-intensive and don’t allow you to increase trust and reliability, at scale.
There has never been a structured way to deal with this in the past, nor has it really been built into everyone’s day-to-day workflow in the past with as much automation as possible.
The idea is that Soda fits in as part of your data reliability and observability stack. Modern software products have an infrastructure layer, which comes with infrastructure monitoring. Then there’s the app code itself which is monitored by application performance monitoring. These two realms have their unique differences, but they have a lot of similarities. The next layer is data, which also includes Machine Learning (ML) models. Those are the four layers that we see.
Data and ML are the two new kids on the block.
What does a “day in the life” look like for you?
As the CEO of the company my day is quite varied, but it’s mostly focused on hiring new talent and ensuring our team members have what they need to get our product and platform into the market. I’m also responsible for setting the company strategy, communicating our objectives to the team, and dealing with everything else involved in building a company, such as talking to prospective customers or partners, and any legal or finance needs.
My co-founder, Tom Baeyens’ day-in-to-day life looks very different - he’s one of the key people that we have driving our open source strategy. He’s currently working on a major new version of our open source project, which will introduce a new language for data reliability (you heard it here first!). He spends most of his day coding and speaking with contributors, users, and customers with the community, figuring out what matters most to them, so we can build an amazing product for them.
What does the team look like?
At the moment we are product and engineering strong - the team is about 35 people. We’re split across a number of domains that are strategic to the product and our customers, and in turn the company.
For example, data testing is a domain for us. Data validations, automated monitoring, incident management, they’re all different domains. These domain leads set the strategy and they have a team to build out that capability within our products.
Our domains have an overarching product manager. The product manager speaks with our customers, analyses product usage patterns, and figures out what’s really important for our community, and then each domain lead works together with the product manager to set the strategy and priorities within that domain.
We’re building out our go-to-market and field team, and currently have roles in business development, sales, and marketing. And we’re also supported by our talent and operations leads.
How do you handle the balance between commercial and open source?
When we started the company, what really made both Tom and myself click was the adoption of our products. We don’t like products that are gated behind a lot of license agreements and not available to everyone. That was the spirit from day one, we wanted to build products that we could put in the hands of everyone.
We started in closed source, so that we could understand all of the different stakeholders that are involved in this problem. What we found was that data quality is really a team sport. There are many people involved, it’s not something you can do on your own. Collaboration is key.
We figured out that what was truly important for data engineers, is that they are able to easily integrate tools into their data pipeline as part of their toolkit. They need to have flexibility in terms of how you test data from different sources and ideally, they can easily add new data sources to test on. Also, they found it important to understand the logic so that they could configure our tools in ways that would deal with massive volumes of data. That was a prime candidate to be open source. Being open source provides a lot of value to our offering as well as our community
The cloud component of Soda is the management and collaboration layer - the workflows, associated metadata, and results. Open source means every individual can start using it today, and they’ll get value from it. The commercial component makes that a better experience for teams.
How did you first get into software development?
I always had a fascination with software development. My first computer was an IBM 486 that I tinkered so much with, that I broke it!
I started mostly with a bit of Java but today when I tinker, it’s really only Python. I also love trial and error, low code experiences like editing YAML configuration files.
I grew up in a rural area, so the software world often felt quite far away, but we were lucky to have a company called Lernout & Hauspie that was in my region. It specialized in speech recognition. I found what they did fascinating and still cherish my experience of going there for the first time. They managed to IPO, but unfortunately weren’t around for very long after.
After my studies I decided that I didn’t really want to be developing and building products as an engineer every day. I wanted to explore the business side, so I decided to go into a role at a software company as a solution architect. I helped implement data management and data management tooling at companies, and evolved into management positions across the business from there.
Is the Soda tech stack all Python?
It’s a combination. For everything open source that touches data, we use Python. We might extend this more into Java, with for example Scala. On our backend, it’s a mix but predominantly Java.
What’s the most interesting development challenge you’ve faced building Soda?
I think the most interesting thing is when we have a problem in which multiple stakeholders are involved. It becomes super complicated to create a great user experience for each stakeholder. It’s already complicated to create an amazing experience for a single type of user. So multiple stakeholders just increases the complexity.
That user flow is an obsession! For engineers, this is typically the CLI - things like clarity of error messages, how logging works, and how the pieces fit together.
For a less technical user, it’s very different. The UX is then primarily in the cloud - easy signup and login, guided onboarding, and getting to value in minutes.
Another thing currently in development is the Domain Specific Language(DSL) we’re building for data reliability and observability. It is something that I think is super exciting and particularly looking forward to its availability.
My co-founder, Tom, built a DSL in the Java workflow space a while ago, known as jBPM. Tools such as Activiti, which was a very sizable project with great world-wide adoption, came out of that. The DSL made it easier to configure the engine and ensure that less-technical users could get involved in setting-up workflows.
We’re applying the same recipe to data reliability and observability.
What’s the most interesting tool or tech you’re currently playing around with?
The most interesting one today in data is dbt. This notion of what Terraform has taught the world of “infrastructure as code”. is just fantastic. Seeing that being introduced in data, creates such an easy workflow.
We also just introduced OpenTelemetry to our own stack so we now understand exactly how users are using our software, which CLI commands they use, why they get stuck and so on. That is something we do using both OpenTelemetry and dbt.
Describe your computer hardware setup
At home, I have a desktop and then I use my laptop on the go. I’m an occasional gamer so I have a gaming machine, and was fortunate enough to do my setup right before prices skyrocketed (thanks crypto!). I have an RTX 2070, a Logitech G502 mouse, and a Ducky keyboard.
For my laptop, I really choose practicality and power, but it needs to be a small device. For the last decade I’ve been using a 13 inch MacBook Pro.
Describe your computer software setup
Email: Gmail, but I’m trialing Superhuman at the moment
Source control: Git.
Describe your desk setup
At home, because I live in a pretty small apartment in Brussels, I don’t really have a proper setup like a home office desk, so I end up sitting around different places.
It was a bit of a privilege during COVID that we had an office. We are fully remote, so we couldn’t do the team meetings at the office anymore, but it did allow me to keep that habit. Some days, I like to be able to just get out of the house and walk there. It makes for a clearer switch between work and personal life.
The office is an amazing building, it’s carbon neutral, and it has great views. It’s a former train station in Brussels where they used to gather all of the goods coming from outside of the city into the city. It’s a massive site, and a really inspiring place to be.
Daytime or nighttime? Night time.
Tea or coffee? Both. It depends on the time of day. Tea at night.
Silence or music? Both. If it’s music, it cannot contain lyrics. I like either classical music or Berlin house or techno. Either work.
What non-tech activities do you like to do?
I love board and computer games, particularly, strategy games like Age of Empires. Outside of that, I’m a sucker for classic cars. I have a 50 year old Fiat convertible. When the weather is good, I love to take it for a ride, be out of the city, into nature, and then go for a hike.
Find out more
Soda is a data observability tool. It was featured in the Console newsletter on 9 Dec 2021. This interview was conducted on 25 Nov 2021.
Subscribe to the weekly Console newsletter
An email digest of the best tools and beta releases for developers. Every Thursday. See the latest email.