AI Visionary Henrik Leijon Joins System Verification as New AI Lead

Can you tell us a bit about your background and how you started working in AI?

I have a background in physics and software development. I started out by studying physics at Uppsala University, but took a (very long) sabbatical to work as a software engineer for about 10 years. At some point I got tired of pure software development and went back to university to complete my exam. By the time I was finished, a PhD position with the ATLAS Experiment at CERN was announced at Uppsala University, which I found to be very exciting. This was in 2011, just the year before the discovery of the Higgs boson. While doing my research, I transitioned from working primarily with software development to data engineering and data analysis, where the software I developed acted as a tool rather than the final product. During this time I was introduced to Boosted Decision Trees (BDT) – a simple Machine Learning algorithm that we used for classification of elementary particles. There was also a lot of buzz about Deep Neural Networks (DNN) / Deep Learning (DL), which were concepts at that time. My interest was piqued, and I have been working exclusively with data engineering, data analysis, and machine learning engineering since then.

What previous experiences do you bring to your new role as AI lead at System Verification?

After my PhD, I started as one of two Machine Learning Engineers at Prisjakt. Together we developed a method to automatically parse offers from shops and index them, reducing greatly the amount of manual work required. We grew the team and I was eventually assigned Team Lead (while also doubling as Product Owner and Scrum Master for a period), and continued to develop algorithms for product similarity, recommendations, price predictions, and sentiment analyses for reviews. After that I joined Spiideo – who produce and analyze video for sports teams – as Tech Lead for their ML team, where I was introduced to Computer Vision (CV) for the first time and discovered the famous YOLO model for object detection for the first time. Just before I joined System Verification I was employed by Sogeti as the AI Lead for Region South. During my assignments at Sogeti I worked as a senior data scientist/data engineer/IT architect, with everything from the Azure Data/ML platform and Databricks to on-prem, bare metal/OpenShift Apache Hadoop/Spark and JupyterLab environments for data analysis. I gained a lot of experience with the three main CV tasks: image classification, object detection, and image segmentation. I also held many education sessions, both introductory and master classes, on a broad range of subjects.

What has been your most exciting project in AI so far and why?

I worked with the Department of Transportation in Sweden to detect cracks in concrete railway sleepers. Detecting cracks in concrete is one thing, but the big question is how to determine if they are significant and require maintenance, and if so in which timeframe. In this project, I both developed the segmentation model and devised an optimization method based on Bayesian statistics to correctly classify sleepers into one of three categories. I also got to act as a general expert on data science subjects as well as Python development and Linux in general.

What attracted you to System Verification and this specific role as AI lead?

What I lacked at my previous position was freedom to discover exciting projects. I really enjoy deep diving into a specific problem and focus my attention on it, but to be happy in my role I also want to meet people from different industries, and both understand their data, and help them solve their problems. I hope I will have the opportunity to experience this with System Verification.

How do you view your role as the first AI lead at System Verification? What are your main goals and ambitions?

My primary goal starting out is to raise awareness and educate. I want to meet both employees and customers and see if I can engage them into learning more about AI, and discover cases where they can employ that knowledge.

Which areas within AI do you think have the greatest potential to improve our services and products?

Automated QA of software is the first thing that comes to mind, including test authoring and code analysis by AI, but also anomaly detection and pattern recognition techniques for deployment in production. A second area I think will become much more important as AI becomes more ubiquitous is QA for data and AI, including everything form securing data quality, model performance, model edge case detection and analysis, MLOps (CI/CD for AI), monitoring, and software QA for supporting applications.

What do you consider the most important factors for building a strong AI culture within a company?

Interest and engagement from current employees and the leadership. And a willingness to try and fail, and try again.

What motivates you the most in your work within AI?

Discovering secrets hidden in data, optimizing tedious work and reducing waste/increasing efficiency. You didn’t ask, but what demotivates me is the hype around LLM:s and the sense that anything is not only possible, but also easy. It is possible to do a lot of things and to do them well, but it might take a lot more effort than leaders trying to catch the AI train realize.

How do you stay updated with the latest trends and developments in AI?

Mostly news, through Reddit or Kaggle, but I often want to understand a topic in depth, so often I prefer to read a few scientific articles per month rather than skimming ten news articles per day.

Can you share a personal interest or hobby that might surprise us?

I’m not sure I have a hobby that surprises anyone. I enjoy learning to recognize wildflowers, in particular meadow flowers. I have an app on my phone that uses AI to classify flowers based on a submitted photo which I use quite a lot during May and June. Me and my wife also go by the No Mow May-principle at our house, so our lawn is covered in primroses, daisies, sweet violets, speedwells and many other flowers during the spring and early summer. Other parts of our yard I let grow into the end of June, and cut with a scythe to help the wildflowers spread.

How do you see the future of AI in your industry and what opportunities do you see for System Verification?

I think we will see a lot of hype, hope, naive experimentation, and failure, as well as some successes. Some companies will want to jump on the bandwagon early from a fear of being left behind, without having a clear idea of what AI can do for them. I think that the companies that succeed will be those that have invested in partnerships or recruited the right competence and built engagement with their partners and employees. I think that everyone would do well to remember that companies such as Google, Meta, Netflix, AirBnB and many other big players in AI have been doing this for 10+ years, and that just because the technology is easily available, the potential advantages may not be. I see a lot of opportunity for System Verification in helping companies ensure the quality of our customers’ data, AI models, and surrounding infrastructure. My experience is that data scientists love working with data and developing new models, while development of supporting software, monitoring, and maintenance of the entire system often come as an afterthought. (Of course there are exceptions to this.)

What are you most looking forward to in your new role at System Verification?

I look forward to learning more about using AI for software quality assurance as well as assuring the quality of AI products. In particular I hope to learn and help develop a systematic methodology for testing and monitoring AI products. MLOps is a fairly new field, and from what I have found, can be difficult to adapt to an implemented product after the fact. I would like to see the same kind of rigorous testing and deployment principles for data and AI products as we have for software products.

Learn more about our AI services
Have any questions? Reach out!