In founders we trust

Christian Byza


Christian, to begin with, please tell us a bit about your journey, what inspired you to dive into the EdTech world, and the path that led you to

As the kids of teachers growing up in the European education system, both my Co-Founder and I were incredibly fortunate to experience widely accessible, free, quality education – and the impact it had on our lives. It is probably because of this, we were able to find our way from Germany to the heart of Silicon Valley more than 10 years ago.

We do not think of as purely an Edtech product in the strict sense. We are in the business of making you feel good about the time you spend because that truly motivates you to learn something new every day.

But first on my path: I’ve been a builder since high school. It started with LAN parties in high school with hundreds of students and after college. I continued to co-found OMR, which is now one of the largest media and tech events and trade shows in the world. OMR also started as an education product. We taught people the marketing basics and that turned from a small event into a company with more than 400 employees today. 

Half way through OMR, I decided to move to Silicon Valley to learn how to build digital products at scale. I spent some time at Adobe and LinkedIn, teaching people about product management. 

Two years ago, I got a call from one of my best friends and he asked me why I would not start an education business again. 

We started with web3 education first, but then quickly shifted to something broader and more casual in education since AI allowed us to have people learn about any topic, in any language. After 3 months, our first alpha product became one of the largest learning platforms in the world (when you count the number of courses and topics). 

This week, our app released on iOS and Android and we are announcing our $3M Seed round. 

So what is Think of us as “Duolingo and Wikipedia meets Gen AI.” Learning in our app is fun. Users can nurture their curiosities, fill their streak, and get inspired by what the community is learning every day. Our app is like a personal study buddy, and we call her "Lumi.”

Lumi is a friendly giant octopus who creates engaging content with generative AI—you can also find our mascot on Tiktok or Instagram. 

Your company underwent a significant transformation from Cryptohunt to Could you elaborate on the thought process and reasons behind this shift? How did the previous experiences feed into the current iteration of the company?

We started with web3 and crypto education, but saw multiple issues. 

One, the crypto markets crashed and our hypothesis that crypto would go mainstream did not turn out to be true. 

Second, especially as a startup we had a hard time scaling content. Prior to the age of LLMs and Gen AI, we were creating 2 courses per week, which we estimated costing us about $1k in salaries if you factor in writing, design, and review. 

Finally, users kept asking us if we could teach them about other topics like quantum computing or ChatGPT, which was still new back then. 

We started to experiment with AI to see if we could assist the creation process and maybe create 10 courses a week. The idea was to take us to a first draft quicker and layer in human work as a second step, but the initial prototype blew us away and we realized that this internal tool could become the actual product. 

We like to speak about this as our “Slack moment” where an internal tool became the actual product. 3 months later, our users had created more than 60k lessons—all with the power of AI, not a single human involved. And we didn’t even work hard to tell anyone about it.

When I reflect on this evolution of the product or pivot, I am extremely proud of how fast we were able to make this transition. We would not have been able to raise money with a web3 only product and the company probably would have been out of business now. 

The opposite happened. We are now a team of 7 and with the support of our amazing investors, we are building something with more value for society than just teaching crypto enthusiasts. 

But we are still fans of some elements of crypto and web3 and also have a plan to integrate parts of that technology into our product. That is also why a lot of web3 investors support us on this journey. 

Successful brands usually have memorable mascots, and is no exception. Can you share the story behind your mascot, and what it represents for your company?

I am a huge fan of mascots. A consumer brand needs to have one. We were definitely inspired by brands like Duolingo. The owl was taken, so we explored different animals like parrots, beavers, elephants, or the octopus. We asked ourselves: which animal evokes the most positive reaction among our target audience, which animal is smart and has some quirky characteristics?

Octopuses are really astonishing when you learn more about them. They can change their gender, they have 3 hearts and 9 brains. They can re-grow their arms or change color and texture. They are probably the most interesting creatures that exist on this planet. It was the perfect mascot for a learning platform that constantly morphs and adapts to the user's needs, thanks to AI.

Naming our mascot involved quite a lot of brainstorming. We wanted something gender-neutral, something that sounded cute and worked in many languages. We also wanted to hint at our magical experience when you create a curiosity with AI. We ended up deriving the name Lumi from “Lumos” from Harry Potter when he uses his magic wand to shine light. We use Lumi and her AI to shine light on interesting topics.

Amongst the team, we have started to speak of Lumi as “her,” but we keep it to each user to decide which gender they want to pick. It’s a nod to the crazy amount of personalized experiences we can craft with Gen AI.

The strategy seems to work out. No matter where we go, people want to take pictures with Lumi, kids want to hug her and everyone has a huge smile on their face when they see us walking around in our giant, extremely clumsy costumes around San Francisco.  

We also gifted hundreds of small stuffed Lumis to friends and their kids, investors and very engaged users - everyone wants a Lumi and we are awaiting another big order.

Given your experience with LinkedIn and now with, how do you view the transformation of education via AI and ML? What potential do you see?

It is interesting. In theory, learning is the perfect use case for AI: Take existing knowledge and recompile it. But when you look at the large players, we see AI being used mostly in very simplistic ways. Many of them just added a simple chatbot. But chat is not a really great product experience. A user does not know what they should start to ask. They can’t infer what the chat bot can do from looking at a blinking cursor. And typing everything in natural language is exhausting. 

We see AI generally as a commodity. That is also how our office neighbors at OpenAI talk about it. An LLM is like electricity or a database. It is just another tool that enables companies like ours to make business models work that were not possible before. 

Think of unit economics. We were at a big ed tech conference recently and apparently, some of the large players spend up to 6 digit amounts for a single course and it takes them weeks if not months to create them. On top of that, almost none of these players is profitable. I heard that LinkedIn Learning makes more than a billion USD in revenue - but is not profitable. 

AI now changes this—if used in the right way. E.g. we spend between 3 and 7 cents for one curiosity (meaning a lesson) that our users create - again, compare this to > $100k per course. And it takes 20 seconds. 

The more curiosities our users create, the more our cost also goes down because users learn about the same topics. 

“What is an indictment?”
“Why do we have wildfires?”
“Teach me about an octopus.”

Curiosities like these are taken hundreds of times and we just had to pay once to create them.  

So, we are at a big inflection point for learning. AI learning companies can suddenly be very attractive business models since the main cost that accrued in the past - for creating content - can almost be neglected. We have a huge advantage over the existing players since they will have a hard time changing their business models so quickly. 

Can you talk about the unique challenges that faced in developing its AI and ML-based educational software? How did you overcome them?

I think it comes down to three things. (1) Making the right hiring decisions (2) having the right creative ideas to solve problems and (3) a lot of luck that we started to get into AI relatively early and being so close to the most relevant players in the market. 

We sit in the heart of the “AI epicenter” of the world. I am looking at the OpenAI HQ while talking to you. Around us (we built a big co-working space for AI and Web3 companies over the last year) are about ~75 other founding teams - all digging into the possibilities that AI will bring. We surround ourselves with smart people, hire the best, and tackle the issues as they come. 

But I can take one example to be a bit more specific: hallucinations. 

If you just read the media, you might think that any content that AI creates is incorrect, is not creative, or that a human can do much better. We fundamentally disagree. We found ways to mitigate most hallucinations almost entirely. We figured out that secret sauce because we are focused on a very narrow scope of AI - explaining things mostly related to general knowledge (For the German readers: “Allgemeinwissen”). We have a unique mal of trusted content, and we have user feedback at massive scale to help us retrain our models for this specific use case.

Large LLMs can not easily do this, because they do not have a focus. But with our specific use case, this is a great way to improve the quality. And we like it this way - if it were easy, anyone could do it. We are working hard on it, but are already running lap two while others are still getting ready to start.

Are there any other problems? Every day there is one, but that is the fun of building a company. You hustle and fix them. 

Many traditional educational institutions are resistant to online learning. How does plan to integrate its AI-powered learning platform into such institutions?

Traditional learning focuses a lot on serving specific government bodies, instead of building something that people are crazy about. Our plan is not to be a tool for 11th graders in a high school in Texas. In fact, we already have users from more than 150 countries today. Remember that AI speaks almost every language; building a global product from day 1 is only marginally more expensive. You just ask your question in Japanese and our AI will create a Curiosity in the same language. 

We also know about the long sales cycles in educational institutions and that selling into those is a very challenging task. This is yet another reason not to focus on that market and tackle it bottoms up first with a consumer focus.

How do you see's B2B approach evolving? And how does it differ from your B2C approach?

We think B2B learning companies get it all wrong and are extremely exposed to disruption. The reason is that their products suck, and employees get forced to use them. And these institutions then complain that nobody wants to learn or people don’t retain information. And these very institutions come inbound to us already. One of the largest tax and consulting firms in the world reached out if we could also provide our product in a corporate learning environment. 

One of the largest army forces of the Western world also reached out with the same question next to many others. 

We do have a plan to eventually build corporate solutions where an admin can upload content and our AI will then create engaging Curiosities around their proprietary content. Employees can learn on the go in an app that is way more fun than anything in corporate learning that you see out in the market today. 

But our strategy is pretty clear. We are building an amazing consumer product first - one that people love and use every day before we jump into monetizing companies.

You are relying on third-party technologies to develop your product. How do you ensure the uniqueness and defensibility of in a rapidly evolving market?

We use different LLMs to power our learning solutions. But we are doing a lot on top of the “out of the box” solution. 

As mentioned, we believe that LLMs or any kind of AI is a commodity. 

But there are many areas where you have to fine-tune and adjust AI models in-house. E.g. We are building a lot of expertise around preventing hallucinations in AI. We just hired an ex-Google and Meta engineer to help us build out that area of the product. 

After all, a learning platform needs to be spot on. If you take a course and see that the content is not correct, we might lose you as a user.

With the increasing adoption of AI in the educational landscape, what ethical considerations should be considered? How is tackling these issues?

We thought about this from day one. And there are different things that we do. First of all, we do not allow certain topics on the platform. We have a long list of topics around violence, harassment or sexual content that we just do not want to teach people. 

Very early in our company, we defined a set of 10 values that should guide us in how we build products and how we think about these topics or speak about them. Two of them go in this direction. “Everything can always be improved” and “We own our work end-to-end” touch on how we own what we do, but also leave room for improvement. 

What aspects of integrating AI and ML into educational software excite you the most? What potential changes or advancements are you most looking forward to?

Like you, my Co-Founder Arndt and I grew up in (rural) Germany. We all went through the German education system where access to knowledge is generally free. 

Historically, we never found that e-learning startups had compelling offers for users. Think of LinkedIn Learning and others that charge you hundreds, if not thousands of dollars a year. Their content is not even very tailored to your specific needs, and they have to charge users a lot because their businesses rely on expensive content production every day. 

With the power of AI, this now all changes. Suddenly, you can teach anyone at almost 0 marginal cost. 

Looking ahead, how do you foresee the impact of AI on the future of learning and education? How does plan to stay ahead in this revolution?

We surround ourselves with the best people in the best place to work on this topic. As mentioned, we built a pretty phenomenal co-working space with more than 150 founders and developers over the last 9 months. One of our investors gave us a 3000m2 office and we brought in a lot of those founders that everyone talks about in the Valley now. We organize hackathons and talent fairs. We sit right next to Open AI to stay even physically close to everything that is happening.  

We are all very optimistic about the abilities of what AI can do for learning. Remember that ChatGPT was released ~10 months ago. We just see the beginnings of what is possible and are excited to be part of this so early. 

Can you talk about the algorithm behind How does it determine the right course content for each individual user?

There are a few aspects to this. We consider us as a social learning platform, so we show you what your friends are learning and push editorial content to our early users. 

We also learn from what you do on the platform and are building a massive learning graph. 

At the end of each “Curiosity” that you take in the product, AI recommends a topic to dive deeper. 

Soon you will also be able to tell us what topics you like so that our recommendations can get even better. 

In the end, most users do not know what they want to learn, so recommending the right content is key to our success. 

What was the biggest challenge in transitioning from LinkedIn to founding an educational technology company? How did your experience there help you?

What blows my mind every day is how quickly we can make decisions. From an idea to having a product in front of our users often does not pass more than a few hours. As a product person by heart (I was a product leader at LinkedIn before), this is what excites me the most. 

You cannot iterate that fast in a corporate setting and there is also a lot of alignment required. I did not expect how much more I would enjoy the fast pace of a dynamic startup. 

But there are also many things I learned from my past corporate job life. Having company values early on, defining and repeating a clear Mission and Vision every week, defining career paths, or writing up a promotion document seem like big company things, but work well in a smaller startup. 

On the challenge side: It was not a hard cut since I started and raised our first million for this company while I was still working full-time at LinkedIn. It was more the other way around that I could not spend enough time at But I am also super grateful how LinkedIn allowed me to explore my entrepreneurial passion and get on this path. 

In a time when deep fakes and misinformation are prevalent, how does ensure the authenticity and credibility of the courses and content on its platform?

We talked a bit about this before but let me dive deeper here. Generally, a language model has the ability to speak and use its endless knowledge to put this into an output that we can consume. Oftentimes, it does not know the exact details or mixes things up. When the model does not know the answer, it starts to come up with random content rather than the truth. That is what we call a hallucination. 

While models like GPT-4 improve its learning function (since that is of course also one of their largest problems), we identified ways to mitigate this even more. 

We are not quite ready to speak publicly about how that all works, but we will soon be able to guarantee content accuracy and sources, which is completely novel in the field of AI right now.

We are starting to roll this out in certain verticals like history or biology and will continue to do so as we grow. 

Lastly, could you share some upcoming developments or features users can look forward to in How do these features align with the future of AI-powered learning that you envision?

We see ourselves as a social learning app. If you use the app today, you’ll notice we try to create an experience that is fun and similar to a game, rather than traditional learning. 

The first gamification feature live today is our streak. In the next few weeks, users will also be able to get points for their learning efforts. These points can be redeemed to get your streak back or to get other benefits in the app. 

We’re also working on a challenge feature where you can compete with your friends. 

Most importantly in our development process is probably how we involve our users from the beginning, understand how they use the app, where they drop and build things that they want to see and have fun using every day. 

In the end, I have a great idea of how you will see what we build. Sign up and use the app and you will be notified about those first:!

Thanks for having me on!

Join Robin’s Newsletter

Insights from founders, investors, and the market.
The best stories from our community.

Robin Capital