Rens ter Weijde, Co-Founder & CEO of KIMO – Interview Series

KIMO is a Dutch start-up founded by two Harvard alumni: Krishna Deepak Nallamilli (India) and Rens ter Weijde (The Netherlands). The team is focused on building the artificial intelligence needed to generate individual learning paths through digital learning content.

While online learning has taken off, it does feature dropout rates as high as 95%. Why is the success rate so low?

When we started KIMO, we surveyed a few hundred users to understand the situation better. First of all, most online providers provide MOOCs (online courses), but users perceive MOOCs as a significant time commitment. They would often use a lot of ‘shorter’ workarounds, like reading articles, listening to podcasts, asking questions on Google etc, which better fit their daily schedule. So learning is more multi-channel in practice than these providers allow. In addition, many users expressed that they missed guidance in their online journey. The result is that they spend a lot of time searching, trying to decide what to study etc.. A third reason deals with the relevance of the actual content. Online materials are often static, pre-recorded, and not fully relevant to them. You could say that the content is not personal/relevant enough for them – or at least not relevant enough to justify the time spent.

Many users claim to be bored and frequently cite a lack of engagement as an issue, why do you believe that users feel disenfranchised with online learning?

I believe there is room to make learning platforms better on at least two main dimensions. First of all, better intelligence is required to guide users on better journeys, and provide better content recommendations. You could say that this is the required R&D for the education sector, heavily related to AI algorithms. The second element is the other side of the value chain: the UI/UX and final experience for users. Most LMS systems are seen as boring and outdated by users. They are far from the polished, real-time, social and personalized software that users expect today.

Could you share the genesis story behind KIMO and what attracted you to solving the online learning problem?

Sure! KIMO started when Krishna, my co-founder, and I met in Harvard Business School. We loved the environment, but realized at the same time the good parts of the experience were not scalable to people across the globe. We decided over a Whisky in a hotel lobby that we could try to create a ‘digital career coach’. That coach was the first version of KIMO.

Can you discuss how AI is needed to generate individual learning paths through digital learning content?

In fact, KIMO relies on a multitude of AI models in the pipeline. Some models are inherently simple, others are more complex. The common thread is that most models rely on natural language as their input data (NLP, e.g. transformer models). These models are behind the content recommendations you receive, the clustering of content into specific topics, or recognizing the critical skills required for jobs. We also have some more experimental ‘generative’ AI models, like the model that answers the content-related questions inside of the KIMO app. Should this work sufficiently, it’s a step closer to the automation of professors that we envision.

Can you elaborate on how an AI system can learn to understand jobs in great detail (e.g. hard skills, or soft skills)?

In simple terms: we decided to ignore existing databases (O*Net, ESCO) for this work as they were not granular enough and outdated. Instead, we built a system that can recognize ~40,000 skills inside jobs in the market in a near real-time manner. You could say that our system ‘reads’ all these job profiles in order to predict what kind of skills are currently required for the jobs. Those recognized skills are then later clustered into soft- and hard skills.

Can you discuss how personalized learning works on the platform, such as how the system will know what type of content performs best for each user such as articles, videos, podcasts, papers etc?

The simple answer is that we match users and content through vector matching, which is common practice in recommender models. The harder part is to decide how those vectors are built up, in other words, what elements are weighted in. For now, the system is relatively simple and works with learning preferences of the user and popularity scores for online materials. The future will be more interesting, as we’re trying to weigh in the current state of the user (e.g. their job), and the desired end state.

What are some of the current machine learning methodologies that are used in the KIMO system?

We use many different models, depending on the task. But I can say that we have a deep love for NLP models that use attention, thus transformer models.

Where do you see the future of online education in 5 years?

In short, I see online education moving from ‘boring, lonely and one-size-fits-all’ to ‘highly engaging, social and personalized’. Online education companies need to escape the idea that they exist in a traditional, slow-moving industry. Instead, they should realize they compete in the age of information curation, right at the heart of many important trends today.

Is there anything else that you would like to share about KIMO?

Yes. KIMO is still a baby, or ‘beta’ as we call it. Download the app, try it out, and send us your feedback!

Thank you for the great interview, readers who wish to learn more should visit KIMO.