#331 The Future of Data & AI Education Just Arrived with Jonathan Cornelissen & Yusuf Saber

Generative AI is transforming industries at an unprecedented pace.

But as AI changes how you or.

Your team work, one thing is clear.

Your skills also need to evolve.

At datacamp, we offer everything you or your team need to adapt and thrive with AI.

Whether it's business users looking to get the Most out of ChatGPT and Copilot, or developers and data scientists looking to fine tune models, you can learn the.

Entire AI skills spectrum on Datacamp. Power your AI transformation today. Start learning@datacamp.com.

I think it's, it's the biggest kind of S curve in, in the history of technology education. You can leverage Genai to build a much more personalized experience that adapts to the learner in, in various ways. We're going to get to the point where eventually education technology will beat the kind of human one on one tutor. And I think that's an incredibly powerful moment.

GPT 3.5 came out in late 2022 and it felt okay. Now there is a chance we could do something different. We latched on the idea of instead of creating content, we create the ingredients that then make up the content and then we have the AI actually deliver the content in real time.

It's not just that the new curriculum will become available, it's also that there's an infinite amount of versions of that curriculum that will emerge over the next year. We are now testing this with kind of real learners and the early results confirm what you kind of intuitively expect, which is like, people just enjoy the experience better. They feel like it's more uniquely for them and it drives engagement up the.

Journey from seeing a particular audience that has particular knowledge needs, going from that to having a course that can be delivered can shrink significantly from months to days. And we think that this can all be possible within 2026. So that's like really, really exciting.

Welcome to Data Framed. This is Richie. Of all the ways generative AI is changing the world, its effect on education has perhaps the biggest positive impact. Datacamp has always been at the forefront of using technology to educate people. We're continuing that trend because datacamp has just acquired the AI Native education platform and Optima. Today we're going to find out what Optuma is and what the implications are for datacamp users, datacamp admins and the future of education. Our first guest is Jonathan Cornelisson, the CEO and co founder of datacamp. I felt like he needs very little introduction at this point. He's been on the show enough times, but suffice to say, he's been building Datacamp for the last 13 years or so. Until a few days ago, our next guest, Yusuf Saber, was CEO at Optuma, and as of now he is Chief AI Officer at datacamp. Before founding his startup, he was one of the top data scientists in Dubai, acting as VP of Data at Taliban, which is kind of Uber eats for the uae. So let's hear out the AI revolution for education. Welcome Josef, good to have you on the show.

Thank you. Thank you for having me.

Excellent. And Joe, welcome back.

Thanks for having me.

Brilliant. So to kick off with what is Optoma? Tell me all about it, Yousef.

Yes. So Optoma is an online learning platform where the content is created by a human expert, but is delivered by AI in real time to the learner.

Okay, easy peasy. So it's edtech, but with an AI twist. Now it feels like almost every low rent datacamp competitor is advertising. Like, oh, I've got an AI tutorial. This sounds different. So tell me, how is it different from a standard AI tutor?

Yes. So in most instances the content is still the good old content that we have been working with. So I'm still learning from a video or by reading an article. So the content itself is static, but on the side there may be an AI that I can speak to to ask a question. Most often it's kind of added to the experience. It's not naturally part of it. And if I don't think about asking a question, then it doesn't really help me with the learning process. But what we are trying to do is that the AI actually delivers the learning as opposed to being on the sideline waiting to be asked the question. It's like when you think of a human tutor, you don't think of someone sitting on the side waiting to be asked a question. It's the entity that's actively delivering the learning. And that's how it is on Optima as well. What?

Wow. So this is like really, it's imagined from the ground up being an AI experience rather than just sort of a bolted on feature. That's very cool. So I'd love to get into the details more, but it sounds like it's gonna have some pretty big implications for education in general. Joe, do you want to give us some wider context on this?

First of all, I'm incredibly excited because I think if you kind of take a step back, it's well researched that the best way to learn is kind of having your own private tutor. And this has been researched over the last few decades. So it's a well established fact. I think Bloom was the researcher that first tested this and kind of compare it sending people through a conventional classroom experience. And then he tested sending people through a one on one tutoring experience. And the people that went through one on one tutoring, they were not just a little bit better, they were two standard deviations better. So what that means is you have kind of somebody who would perform average going through a classroom experience and they would get into the top three percentile after one on one tutoring. I'm saying this because one on one tutoring has always been the best kind of experience, but it's very expensive and it's just not affordable for most people. I think with Genai and with some of the technology that Optima has created and that we'll continue to invest in, we're going to get to the point where eventually education technology will beat the kind of human 101 tutor. And I think that's an incredibly powerful moment. So kind of similar to how you now have Tesla and Waymo who are building the AI driver and that AI driver is starting to become better than a human driver. We're going to reach the point in education technology where the AI educator is really better than the best one on one tutor. We're not quite there yet, but I think this is the start of that process and I think it's going to be really exciting next couple of years as we build that technology and kind of roll it out into the world.

Yeah, certainly the idea of having a one on one tutor, I mean this makes a lot of sense, that every sort of rich family, they give personal tutors to their kids in order to help them get into university, whatever, do better in school. And so the idea of bringing that to everyone, regardless of how much money you have, that seems a pretty huge game changer for education.

Huge. Absolutely huge. I think it's, it's the biggest kind of S curve in the history of technology education. Right. Like if you go a step, take a step back. I think the first wave in online education was the massive open online courses like think of Coursera, edX, Udemy, the video based platforms. And I think what they did is they bought a conventional classroom online and I think they deserve a lot of credit for doing that because there's some amazing content that came online as a result of that first initial wave in online education. But they all share the same problem, which is like you're just passively watching videos and so it's not personalized. And then Datacamp came about and we kind of as A counter reaction. We said, hey, learning has to be active. People learn by doing and has to be engaging. People actually have to work with the technology, solve challenges. And so we kind of wrote this second S curve in online education. We were not the only ones. I think Duolingo, Codecademy and a few others were ahead in that wave, if you will. But now today, I think we're at that third S curve where you can leverage Genai to build a much more personalized experience that adapts to the learner in various ways, it adapts to their pace and so on. And I think what Yousef and Optima team have built was kind of the best incarnation I've ever seen of that kind of vision, of the early phases of that vision kind of materializing.

It's a bold claim, having an AI tutor that's sort of as good as, or maybe even better than the best human teacher. So definitely something aspirational to work towards. Exciting stuff. Okay, so let's get into more of the details of Optuma. So, Joe, you alluded to the idea of personalization. And before Josef, you were talking about how the AI, it generates exercises on the fly. This seems very cool. Talk me through. Why is this a good thing?

Yeah, so just to make sure that the process itself is clear. In traditional learning, when I'm watching a video, what I'm seeing here is like a set of elements. So there are some concepts or techniques that are being explained, there are some examples that are being given, there's some exercises, there's some illustrations with figures and so on. And all of that is interwoven into this narrative that I'm going through. But that narrative is static. So what we do is say, let's roll this back. Let's extract these elements. So the set of concepts and techniques that are being taught and then all of these components. So we have exercises, examples, illustrations, figures, and so on. So the human course creator kind of prepares all of those, and then this set of ingredients goes to the AI tutor. So instead of, for example, I get the ingredients and then I bake the cake and I give it to you, I prepare the ingredients, and then I come to you, and then we bake it together. So that's essentially what we do. So instead of passing the static already written content, we pass all of the ingredients, because thinking through the ingredients takes a lot of time. It needs, you know, until now at least, needs a lot of deep human thought. And that cannot be replaced with AI. But the act of that interweaving, the delivery, that can be done fairly well with today's AI technology. So elements get to the AI and then it also has the profile of the learner. So essentially, you know, just essential pieces of information that are incredibly valuable to deliver a high quality lesson to the learner. So one, for instance, what is, is their role in industry? Like, what's their current situation? And then second, what are they working towards? What problems do they really care about? So by knowing these two things, like their situation and what they care about, they can then deliver the information to them in a way that is really connected. So they get examples that are from their domain and industry. So they instantly make sense when motivating why the learner should learn a particular concept or technique. It connects it to the problem they're trying to solve and then instantly they have reason to care because they see how knowing this can help me with the problem that I'm working on at the moment. So that's on one side, so connecting the learning to their context, on the other side, connecting it to their ability. So we've all been in classrooms where it can be really boring and we kind of zone out or we are struggling because we, you know, zoned out for a few seconds and then we missed a key point and then now nothing makes sense. So data here will do everything possible to prevent any of those two negative outcomes. So if the learner is struggling with something unclear, it will slow down, make sure to fill those gaps, and only once they're clear, it will move on. So either the learner will express their confusion directly, so they'll say, I don't really get what this means, or before moving on to the next point, they will test if they captured what was just explained, especially if the next point will build on it. And if they are unable to perform the exercise, it will then spend some more time before moving on to the next point on the other end. For people who captured the information quickly or happen to know some of it already, then it can move really quickly through those points. So they are not sitting through content that they already know. So these are like both of them, where they're connected to context gives people reason to care. So they're engaged and adapted to their ability. Then they don't, they're not bored, they don't zone out, or don't struggle and kind of decide that they're not good at math or all of these stories we know that's very cool and I.

Like that there is still a human involvement in deciding what to teach. You are making use of like human Experts. But then it's just in the delivery layer. That's where the AI works, because AI is good at changing tone of voice or other things like that. So subtle changes in how to explain things. Your point about making people care about the context reminds me. One of the early courses we made on datacamp, it was a Python course, and it included a data set on Galapagos finches. And this was like a really famous scientific data set from, like, Darwin, and it was like, crucial to his theory of evolution. It was like, safe choice. And we got overwhelming feedback. Every student was like, I don't deal with finished data. I'm trying to add a click through rate or whatever. And everybody hated this. And so actually, students really need something that's connected to why they're learning, whether it's related to their job or something like that, to actually make them want to do this. So I like the idea that you are worrying about what students care about. So are there any other benefits around personalization? I mean, you mentioned a few.

Yeah. So I think that the one that I always go to is when people have a reason to care, they're engaged, they pay attention. So I think by connecting to the problems that they care about, then that's kind of the highway to engagement, is to show them how this can help with the problem that they are currently focused on. That's one. Then two. I think all the time people move on while they haven't really understood. And then that gap can build up, and then quickly they stop learning because they get this impression, I'm not good enough for this. And that's really sad because they are good enough for it. They just missed some building block, and then everything that builds on it is suddenly like, unintelligible, and they can't continue. So by preventing those gaps from forming as early as possible, we can make it a lot less likely that someone will give up for no real reason just because they missed the point. So I think these two reasons are, I think, pretty powerful, like motivators for delivering content in this way.

Absolutely.

Yeah.

I mean, engagement is such a huge challenge when you're learning anything. Like, there's always something better to do or a different deadline, or maybe stuff is too easy or too hard, or there's always a reason to give up. And making sure that people actually continue is very important.

Totally. And maybe to link this to, I think Andrej Kravati recently talked about this on another podcast, kind of talking about how it's much easier to stay motivated if you have your private tutor because they adapt in so many ways and it solves one of the fundamental problems I think online education has had to a large extent, where engagement has always been kind of the key problem to solve in online education. And then if you look at companies like Duolingo, that from a financial point of view are doing really well. But you could argue from an educational point of view, they've essentially chosen the route of saying, hey, we, we're going to gamify the experience and kind of move away from learning and move towards gaming, but in a way that doesn't necessarily benefit the educational aspects. And I think what this technology ultimately now enables is to drive more motivation, but in a way that is aligned with better learning, just like an individual tutor would do. And that's the first time that's possible. It's something even at datacamp, our engagement has always been way higher than other kind of passive learning platforms. But at some point you hit a ceiling where you have to kind of make it easier and gamify it more and you have to start sacrificing learning if you want to increase engagements. And we've always tried to tie that line because our focus is ultimately on making it as effective for a learner. For every hour you spend on datacamp, making that as effective as possible, but then maximize engagement within that. I think with the new technology and what Yousef has built, you don't really have that trade off as much anymore. You can actually make it more relevant to the learner, to their goals, which is more engaging but also more helpful from an educational point of view. So I'm super excited that these two worlds are kind of starting to go inside now, which wasn't true two years ago.

Yeah, absolutely. And certainly I like gamification in general, but if you're just doing some education because you got a passive aggressive message from an owl and you're not really into it, then you're not actually going to learn anything.

It's great, it works, but there's limitations. There's a point at which you actually want to drive engagement by making the learning more relevant and more effective.

Absolutely. Wonderful. I'd like to take a step back and see how we got to this point. So, Yousef, can you tell me, why did you build Optima to begin with?

So I've always enjoyed teaching. It's something I've done since I know middle school. So I would just start these small teaching activities and try and help people somehow through it. It's actually how I met my wife. It was while teaching a Coding class to business students who wanted to learn coding. And in almost every organization that I worked since, one of the things that I was on the side was start some sort learning setup where people could come and build data skills whether and to support their work or to transition into a career in data. And one thing, we almost always used Data Camp as one of the tools for the learners. So it's amazing to be here. I guess all through it it was like small scale. And I was wondering, I'd love to build something here, but I couldn't see how different would that be from what's already out there. Like, there was no, like, yeah, we could build content that's, you know, slightly better in some ways, but there was like, it always felt it wouldn't be adding much. Like there's already, you know, Data Camp that's doing a great job. For instance, on content, on interactivity. There is for a lot of other areas, you know, I could go to Coursera, I could go to Udemy, and there's like plenty of stuff out there. And we thought maybe it's like the physical aspect where we could interact with people directly and kind of teach in that way. And for a long time we left it at that. Like, it would probably be something that we do in person where we can do something that's unique and different. Until GPT 3.5 came out in late 2022 and it felt okay. Now there is a chance we could do something different. So we started working on Optima a few months after that. And the idea was, can we build something where the learning can connect to someone's context, what we just discussed, and adapt the pace to whether they're following or they're struggling? So that was kind of the problem we set out to solve. And for the first nine months, we tried four versions of three versions of Optima. This is the current one, is the fourth, and they're all like a step towards that, but it wasn't still quite right. I think we're still too hung up on the classic way of doing things. So we still had AI on the side essentially. We tried to have it do more and more, but I think the YouTube Coursera model was too strong to beat. At least it had too much of grasp on our imagination. And then it took us a while to kind of reimagine things very differently. And that's where we latched on the idea of instead of creating content, we create the ingredients that then make up the content and then we have the AI actually deliver the content in Real time, because that's the only way. Static content just doesn't leave any space for us to personalize or to adapt the pace. So the content has to be created dynamically for us to be able to do these things in a meaningful way. And it took a few months to get that to work. I think today's technology is pretty powerful. I think there is so much more that we can do with it, but it doesn't kind of give that up easily. Like, people have to work really hard to try and make it work. And it's so new that I don't think we really understand how to use it to maximal effect yet. So for quite a while, it was not reliable enough to actually have actual learners learn with it. And this changed in February of this year with the release of Cloudsonnet 3.7. And for the first time, it actually worked. Like, we could have a learner learn for an hour without any, like, serious glitches. It was amazing. And we felt, okay, we have something here. Let's try and put it out there. So we had a few really great organizations that kind of saw the idea and wanted to be part of it. And then we started having their teams kind of learn on it and adapted to their feedback. And it evolved fairly quickly over those first few months. And I think it was maybe three or four months in after that inflection point when we met. Yeah. And the rest is history, I guess. That's cool.

Yeah. And I love that it started with you teaching groups of people using Data Camp.

Yeah.

So brilliant. And the fact that you had the idea first and the foundation models, the underlying technology wasn't ready, but in the last sort of few months, it's now hit the point where it is a viable thing. So, yeah, you are ahead of the cutting edge. All right, so you mentioned the two of you met. I'd love to hear more about this story. So perhaps, Joe, can you tell us, how do the two of you end up meeting?

Yeah, from my perspective, it felt like very coincidental, and especially the timing, I think, was kind of really fortunate. I'd never been to Dubai, where YouTube and team is based. We have one of our clients, Emirates, who was organizing a conference. And I was going there essentially to meet with them and to meet some of our government clients in the region. And I remember there was one specific meeting, I'm not sure I'm allowed to say which company, but one of our clients there that historically was very happy with datacamp, they started grilling us and. And they started asking all These questions of, like, why isn't it more personalized? Why can't we customize this a little bit more? Why isn't it adapting to the learner? And so we started talking about, okay, what should the vision be? How can Genai be used? And I was very aligned on their vision, but in the back of my mind, I was thinking, I agree with them, but nobody's been able to make this work yet. And so it turns out that company that we were visiting, one of our clients, it turns out they just had seen a demo of Optuma, and so that got them really excited about the kind of AI learning engine that Optima has created. And so one of the people there mentioned, hey, there's this founder also in the education space. Maybe the two of you should meet. And I didn't know. I didn't actually connect the dots at the time, so I didn't know that's why he suggested for us to meet. So almost didn't take the meeting. But the last meeting I had right before I left Dubai was with Yusuf. And as soon as I saw the demo, I was like, whoa, this is incredible. This is like, not to sound arrogant, but, like, I've rarely seen product experiences where I felt like, hey, this is better than what Datacam currently has. And this was one of the first times where I was like, oh, wow, this is actually better. Like, this is what we aspire to build, but here's. Here's a team that actually has already made it work and has created this. This kind of magical experience that adapts to the learner in so many different ways. So as soon as I saw the demo, I was like, wow, this is incredible.

It feels like there were just like, so many layers of chance happening there. That is like, kind of a miracle. The two of you got together.

Wonderful. Let me maybe add a couple of layers. So two days earlier, I had had a surgery, and there was a chance that I wouldn't make it. So I was going to stay in hospital for two days, but with a possibility of extending by a third. So if it is extended by the third, then I would have missed it because it was in a different city. So it was in Abu Dhabi and we were supposed to meet in Dubai.

That is crazy. I didn't even know that.

But luckily I went out on time and it was the only meeting I took because I had to stay because of the surgery, just at home for the next two weeks. So I only came out to have this meeting, and I'm really glad I did.

Yeah, that's that's absolutely fantastic. Good job. You actually made it out hospital to take the meeting. Okay, cool. So, all right, I'm excited about Optimum. Now how do I get access? Maybe Joe, do you want to talk through.

So as of today you have our introduction to SQL course, our introduction to AI course. There will be many more courses and every free user, every subscriber has access to those courses. If you're one of our enterprise and corporate clients, you can get access as well. But you should get in touch with our team. All our B2B clients will be getting access reasonably fast. But not everybody is as excited and open about AI. There's compliance questions and all of that. So we're doing a gradual rollout towards our B2B customer base in the next few months. I think by early next year we should be position to open it up for all the clients that are interested in kind of getting access. But for now for individuals, you can just go to website and look at our introduction to SQL. Introduction to AI. There's many more courses coming and they'll be going live in the next few weeks and months.

All right, so for individuals, just log into datacamp, start taking the courses straight away. Brilliant.

Absolutely.

And I guess if you're a business account and yell at you customer success manager and they'll sort you out.

Yep, yep.

All right, wonderful. Cool. So I'd love to know a bit about like what courses you're building out. So what curriculum areas are you working.

On at the moment?

Yousef from the beginning we've been focused on data and AI. So the first areas we are getting out are the core data skills. We're focused on data manipulation. So this is, you know, how to manipulate data with SQL, with Python Pandas and with R. So these three. The second area that we have invested a lot in is AI engineering. So if someone who is a software engineer or in data science or analysis and wants to transition into being a data engineer, we've built a program. So it's a series of six courses that they can take to transition into a capable data engineer. And we kind of took our best learnings working on the platform and we pour that into these courses plus of course the state of the art techniques that people are applying. So These are the two areas that we will be focused on during Q4 and early Q1 and then beyond that we are still in discussions. But the area that we are most likely going to focus on are AI literacy and classical data literacy. So these are focused on, you know, a primarily non technical audience. And we think that this new approach to learning can be really beneficial, especially to those because the variance in their background is, is huge. So you have people who've, you know, never seen a spreadsheet before or have people who have worked, you know, with spreadsheets for a long time, just have never done like measurement and metrics and analytics before. But they are quite familiar with, you know, accounting, for instance. So this can adapt quite a bit. So the literacy like to think of it can become a lot more interesting and a lot more relevant to the person when delivered this way. So at the moment this is kind of what we are thinking about.

All right, that's cool. I mean, yeah, certainly SQL, Python R for data analyst, data scientists, AI engineer might be a little bit of a hot roll at the moment. I can see where you're building out curriculum and I do love the idea that personalization is incredibly important for data literacy because yeah, you're right, there's so many different people, regardless of their job title, they need some kind of basic data skills. But how you explain that to them is going to vary a lot. So, yeah, an interesting area to watch.

Maybe. One thing I want to add, because we didn't get to that yet, but I think as Yousef explained, you're kind of disentangling the content to learner experiences from kind of the ingredients, the examples, the theoretical content and so on. And that enables to kind of improve the experience for the individual learner and make sure it's relevant to their context, to their skill level and so on. One thing we haven't talked about much, which I think is another benefit that is not yet working, but will be working at some point next year, is you can now also personalize and customize this to specific organizations. If I'm a large organization and I have certain guidance towards my employees around say, AI literacy, certain AI policies, certain technology choices that we as a company want to make. The fact that you disentangle the ingredients from the delivery of the content now also enables you to customize that experience for learners at organization A versus learners at organization B. So I think there's so much more that will become possible because of that disentangling of kind of your core content and the delivery of the content that even for the existing curriculum, where we're moving over some of the popular data camp courses or some of the new AI literacy courses, it's not just that the new curriculum will become available, it's also that there's an infinite amount of versions of that curriculum that will emerge over the next year. So that's maybe an important thing to stress because historically, when you talk about launching courses, people still have this mindset of, oh, it's a static thing that's kind of the same for everyone. You're kind of launching swim lanes where then there's a whole lot of different versions of that course that will emerge. So I just wanted to stress that because that's kind of a new way of thinking about the contents.

Absolutely. And I think for the B2B corporate learning, this is incredibly important to understand which bits are the same and which bits are different, because I think a lot of times when you've got an organization, they're like the learning development team's like, okay, I need to train a thousand people on this particular thing. I've got these learning objectives that I want people to hit. And so the learning objectives are going to be the same for the course. Yes, but it's just the way those learning objectives are kind of fulfilled with the way it's explained, that's going to.

Change from learner to learner. Yeah.

Okay.

All right. So actually on that note, do you want to talk me a bit through a bit more like what the. What the experience is for companies who are going to be using Optuma? Is the admin experience going to change? Is the learning experience going to change on a corporate level?

So in the short run, not that much will change. And then early next year, we're going to start rolling out the AI Optima experiences for all our B2B clients. But the admins will stay in control and they will be able to decide do they want people to see that Optima experience? Are they not ready yet? And the important thing I want to stress is that for some time we're going to have a dual content creation engine where we'll create courses in both the old datacam formats and as well as in the new Optima format, AI native formats, if you will. And so for a while, we'll have that kind of duality. That being said, I do believe the kind of AI native experience or the Optima experience is the future. And if you think about all the benefits from an L and D point of view or kind of a data leadership point of view, they're only going to grow over time. So to give you a few examples, you can set certain goals for the organization and the AI tutor can become aware of these goals. So just like you have a profile of a learner and the AI tutor is aware of that learner and their goals The AI tutor will also become aware of the organizational goals. It might become aware of the tech stack of that organization, it might be aware of the AI policy of that organization. And so you'll have not only personalization for the learner, but you also have customization to the enterprise. And you just can't do that in kind of the old world. You have to enable AI to reap those benefits. Another example that I'm personally really excited about is you can create a content once and these newer LLM models especially, they're becoming pretty good at many languages. So now you can have the delivery in 5, 10, maybe 100 languages once the models get good enough in those languages. But you only have kind of one core piece of content that gets created and it's automatically available in so many different languages. So while we'll give the admins control and we won't push this on anyone, I do want to stress that I think we have to work with our clients and have to figure out how to enable this because it will be very powerful for a lot of corporate learning programs.

Certainly lots to take in and lots of very exciting features, I guess, coming soon. So I love the idea that you've got this sort of corporate level context. Oh, we're doing fancy context engineering here as well as like the personal level. But I guess the headline takeaway there is that the 600 courses on datacamp, they're not disappearing overnight, they're still going.

To be there and we'll continue to ship really high quality content in the old datacamp experience for the foreseeable future at some point because we're still innovating here, we're still figuring this out. At some point we'll flip a switch and we'll go towards the AI native experience first. Maybe the last thing I'll say, because we're a very data driven company in general. It's part of our DNA, part of our culture. It's not only what we're preaching, we're not just bitching here that this is better. We are now testing this with kind of real learners and the early results confirm what you kind of intuitively expect, which is like people just enjoy the experience better and they feel like it's more for them, uniquely for them. And, and it drives engagement up, it drives conversion up to kind of payments and so on. So in the data you'll see the improvement as well. And this is an early version, so I think this is going to get way better over time.

Okay. Yeah, certainly customer satisfaction metrics, incredibly important. I Hope we're paying attention to those. So one of the recurring themes on Data Framed is that one of the challenges of enterprise AI adoption is around data privacy. Joe, you mentioned that all the AI features can be turned off at a sort of group level for corporate accounts. Yousef, do you want to talk me through what sort of personal data is being used by Optuma and what are the privacy features that you've created?

Yes, absolutely. So we have done quite a bit of iteration to understand what pieces of information about the learner truly make a difference. And we realized that it's actually pretty few. So with three pieces of information, we can deliver the objectives that we want. The first one is just understanding which context do they come from, so their role and their industry. So that already gives so much information. And just knowing that this is, let's say a data analyst in the aviation industry, for instance, is already like very, very powerful. Second is what are they working towards? So any idea on like a challenge that they are tackling? The, the kind of like, you know, someone in marketing who's working on, let's say user onboarding, that's already like very, very useful. And third is something about their background. So they, they have studied statistics in, in school or they're doing this course in data in Python. But you know, they, they did R a while back. So by knowing that we can, when we're explaining a new concept that's challenging, we can say, oh, that's equivalent to this thing that you know from R. And then suddenly it makes a lot of sense. So just by knowing these three pieces of information, we can do most of what we set out to do. So that's what we look for. Anything in addition to that may be valuable and that can also be added to the profile, but we don't ask for it. So we keep these three points completely randomized. Like we avoid keeping names of particular projects or clients or companies and the learner's first name. So it makes the conversation a lot more kind of reachable. When the tutor knows the learner's first name and it can appreciate them for their achievements using their name, people have consistently found this to be a lot more relatable than when it's just arbitrary. And so that's kept in only one place. It's an encrypted for profile of the learner. And when someone, let's say, wants to remove that or they have access to it, they can go and see it at any point. They can remove it if they choose to. So that's completely under the control of the learner. For organizations, they can provide bulk context for their group. So they can say, we have these 100 people. This is kind of a sheet that we've exported from one of our HR systems which has some essential information about them that we think will be useful. We can work with that, but only if the organization is willing to do that. It kind of speeds up the process. It can already start with quite a bit of context and doesn't have to ask the learner for it. But that's only if the organization is on board with sharing that.

Okay, so it sounds like there's actually relatively minimal personal information in there. I guess if you're really worried about the AI knowing your name, you can always lie about what you're called. But I do like that. It's a nice personalization feature there.

Okay, maybe to jump on that, because I think there's two levels for this. I think there's one level is what you just described, which is you want to kind of minimize the kind of sensitive information that we're storing while maximizing the impact it has on learning. I do think there's. As this space evolves, there's going to be a question whether the kind of deployment model of learning platforms changes. Because if you think about static contents that's not very customized, not very personalized, the kind of delivery model is you have a central platform that everybody gets access to. And it's different from say if you want to deploy Snowflake, where, yes, that's going to happen on your own infrastructure kind of cloud prem or single tenant at the very least. That's not really been something that you've seen within the educational technology space that much. And I think there's an open question to whether eventually, as you get more personalization, more customization, whether that kind of deployment model changes from kind of a multi tenant to a single tenant setup. I think the jury's still out and we're going to be working with some of our largest clients to figure out what they're comfortable with. Because even if we kind of minimize the amount of information that we gather, there's a question on, at what point does this cross a threshold where people are no longer kind of comfortable with that information being out there?

Okay. I mean, well, I guess like the extreme version of the personalization then is you're learning, you're doing exercises against like a real company data set.

Exactly, yeah.

And at that point you want it.

So the extreme version is you have absolute customization and personalization and you Have a version of this that gets deployed within the infrastructure of a large organization. You're essentially having your personalized, customized learning experience on top of the actual tooling that organization uses. I think that can be very, very powerful. But it will flip the deployment model completely because nobody's going to be comfortable with that if it's kind of all that information goes back to kind of optimal servers or datacam servers. And we're at the very beginning of that kind of trend. So I think the jury is still out on how that shakes out, But I think that's what this technology is ultimately going to enable. And I think there's definitely going to be a lot of interest in that.

That's a very cool future. I feel like we're creating some big tickets there for the engineering team today to come there. Some big infrastructure changes.

Yeah.

Hopefully they're not listening to this episode just yet. Okay. All right. So stepping back from the technology, it feels like this is going to change a lot the way you run a big corporate learning program. So I'm wondering, like, how do you approach this differently, Yusuf, do you maybe want to take a stab at this to begin with?

We found this quite a bit when we work with organizations. So first is who is keen on learning. So people who are generally interested in being part of the learning program and picking the right courses slash programs for them to be part of. I think once they start and they are, you know, part of the program, there is a whole wealth of depth and understanding that someone running the program can. Can benefit from. So before I could just see like the progress or kind of how far they have come, but now if I want to understand, like, why did certain people drop? So I can get understanding of were they confused about the content from the nature of the questions they ask. So when learners ask questions, these questions get labeled. So we know. Is this a question where someone is curious and digging deeper, or this is a question that someone is confused about the content and not fully understanding and then is following up so they can get a sense of why is someone not continuing with the learning? They can see how they like during this. The way the learning happens is a conversation. So the learner needs to do their part to contribute to it. So the tutor will deliver a piece of knowledge, will give them something to work on or to think about, and then they need to kind of respond and then the next turn will start. And it's kind of like a turn based game where they are playing together and a learner who is engaged, their responses make it very clear learner who is just, you know, passively trying to move this along. That's also very clear. So as someone who is running the program and invested in its success, I think they can get far more valuable information and then they can act on that by having conversations with some of these learners, by kind of thinking through the goals and motivations that people have for continuing with learning or by maybe rethinking what people are learning and whether that's directly relevant to them. So I think that can be quite helpful in making these programs succeed.

I love that you got these richer data sources now because I guess traditionally it's like someone either passed an exercise or they didn't. Maybe it took them a few attempts. You got a little bit of data on like how much content people are completing, but the, the conversational data is going to be an even richer source for admins so they can make better decisions about their learners. Actually. Joe, do you have anything to add on that on the way you approach running a training program?

I think we're at the very beginning. But just to build on what Youssef just said, I think maybe an easy mental model is imagine you're kind of a large financial institution. You're looking to upskill 100,000 or maybe more learners. What this eventually looks like is kind of the equivalent of those 100,000. People now not only have a private tutor that will work with them to get them to the learning objectives, but from the organizational point of view, they have a way to ask questions to all these, like, say there's 100,000 private tutors. These 100,000 private tutors can now report back and say, hey, this is kind of what we're hearing. This is the type of stuff people are struggling with. This is the skill level of this group versus that group. And so datacamp historically has built out assessments and it's part of these learning programs and they're very heavily used. And I think that's still going to be very valuable. But the world we're moving towards is a world where you're going to have so much more information and that can help you kind of be way more effective in kind of driving change within organizations. And the biggest challenge with kind of these new technology based and AI is no different, is usually the human components. And it's not only the skills humans have to acquire, but it's all the struggles around the skills. And I think the benefit of the system is it's going to give the people that try to drive that change so much information. On what's working, what's not working and how they can improve their programs. I know that's a very high level, vague answer, but I think to some extent we're still building this out and this is yet to be played out, but it's going to be super powerful.

Absolutely. I mean, certainly whenever we run learning and development webinars, there's all these big questions around, how do I demonstrate that I've got return on investment from the training program? Because every L and D person they've got justify the training to their manager. And having that extra data in there is really going to help with that.

Totally. And that's actually a great example. Like all the tutor will know. The tutor will know why somebody's on the platform, what projects they're seeing value in. And this goes back to the earlier question on kind of what data should we store and how kind of to think about security and privacy because there's going to be a trade off there where the more information we store, the more valuable that's going to be for especially these larger organizations, but also the more risk. And so we have to change our security posture and we have already changed our security posture to kind of take care of that data in a very responsible way.

Wonderful. So much exciting stuff coming down the pipeline. I'm looking forward to it. But can you tell me what are you most excited about in the world of AI Yousef? Do you want to go first?

I think the more powerful the models become, the harder the problems that we can tackle. So I think the point that we started with in the beginning is that we're still quite a bit of a distance from competing with the best human teachers. Just that discerning eye for where the learner is and what they're dealing with and ability to fill the gaps that they might have. I think the better the models become, the more of that that we can approximate. So this year I think has given us some pretty powerful leaps. I think it's just very hard to imagine that cloud 3.7 was just out this February and now we already have 4.5, which is significantly more powerful. So you can only imagine how powerful the models we're working with would be in early next year. So that next wave of models, I think will be what it takes to breach some of the harder problems. So now we're mostly focused on the delivery, but we have two more frontiers that we want to be working on. So the next one is we call it instruction design. So imagine that we've already decided on what are the concepts and techniques that should be taught in a course. But then we spend all of this time thinking through the examples, the exercises, the designing the data sets, and the problems people will work practically on. So all of that, the concept things have already been set, they're right there. And we're just thinking through what are the right elements that we need to put together to support the delivery of that. So I think that's a problem that can really help with. So far, it's mostly been assessing this. So the human content creator still doing most of the thinking, most of the work, and delegating some tasks to the AI. But we think that within early 2026 should be at a point where we kind of flip the roles. So the AI would, the instruction designer would get the concepts and techniques that have been designated for a particular course or lesson and then it would do the work to design all of the necessary assets from examples, analogies, exercises, and so on. And then the human creator would verify those and give their feedback and they would iterate once or twice. And then the final frontier is like the curriculum design. So what are the right concepts and techniques that should go in a course so that it fulfills the designated set of objectives for a particular audience? So now as a course creator, my part would be to define the audience very clearly as well as the learning objectives. But then my AI curriculum designer would scour vast amounts of knowledge to identify the right set of concepts and techniques to include in the course. And then the human course creators part would be to vary that, give feedback, and then the course can proceed from there. So then the journey from seeing a particular audience that has particular knowledge needs, going from that to having a course that can be delivered can shrink significantly from where it is now in the order of magnitude of months to days. And we think that this can all be possible within 2026. So that's like really, really exciting.

That's very cool. So you kind of, I don't want to say solve, but you get in there with having solved the learner experience and now it's like, can you solve the teaching experience? Certainly, like I've created. And of course, just finding the right data set can be a challenge. Like it's got to like have like the right features in there. It's got to show something interesting, you've got the right license for it, and it's got to be like relevant to enough learners. So yeah, that's a big challenge. It can be a huge time suck. So just being able to automate some of These sort of things is incredibly important. Cool. All right, Joe, what are you most excited about?

A couple of things, I think. First of all, I agree with what Youssef said is going from reimagining the delivery to also reimagining the content creation, I think will be super powerful. Something we didn't touch upon yet that I think actually is really important too is if you look at the history of interactive learning and the history of datacamp as well, it's like to build these active learning experiences was kind of customized to a specific vertical. So datacamp was kind of doing this for data science and analytics and we built kind of specific interfaces for a vertical. Duolingo, did it for language learning, codecademy, did it for web development. But I think what's important to kind of realize, if you take a step back is a lot of the technology that we're developing now and a lot of the technology Yusuf was talking about is really generic and it will work for any type of learning, for any type of topic. And so what I'm really excited about is obviously in the short run, AI skills, AI literacy, AI engineering, AI skills for people in different functional roles. That's our priority from a curriculum perspective. However, in addition to that, I'm also excited that we're building a system that, that can focus on active learning, focus on learning by doing for a broader range of topics. And I think this technology enables us to build out really high quality education for any type of continuous learning within a professional setting. And that's what I'm super excited about. Our ability to kind of go broader in the medium to long term as well as a result of this technology. And yeah, I think there's tremendous potential because in a lot of verticals there actually isn't a platform that does learning by doing. And so there's only kind of passive video based education in a lot of these verticals. So the delta with what we're creating here is huge and will yield like much better educational outcomes for sure.

So we're going from Data Camp to everything camp.

Yes. We'll have to think about the name of the company.

Get registering those domain names.

Yes, yes. No announcement on that front today. But maybe one other thing I want to circle back to because I'm very excited about that, is the ability to deliver learning in many different languages I think is also very powerful. I think I live in the US now and you kind of forget how many people in the world would benefit from education that's not in English. I think the number is in the billions. And, and so I also feel very strongly that this is another kind of democratizing force because you can build really high quality experiences, really high quality education once, and you can now scale it to way more people. And so I think that's the beauty of the advancement in technology, is that it kind of continues to democratize the best kind of educational experiences even further.

Oh yeah, certainly. I mean, with translations, the it's been possible to create content in every language or at least the top 50 most popular languages, but it's still a little bit fiddly and technology is getting to a tipping point where we can sort of make that a reality.

It very much feels like it's at that tipping point. My mother tongue is Dutch and there's only 20 million people in the world who speak Dutch and I've been testing those models in Dutch and they're surprisingly good. So if it works for languages the, that are only spoken by 20 million people, my assumption is it's going to work for a lot of languages around the world and that's super exciting.

Yeah, maybe we'll even get like data camp in Klingon one day. Nice. So Yousef, can you tell me what areas of the curriculum is your team.

Focusing on at the moment? Yes. So we are starting with two areas. So first is one core classical data science area which is data manipulation. Data manipulation in SQL, Python, Pandas and R as well as AI engineering. So for AI engineering we developed a program which probably about six courses where we have kind of poured all of the learnings we've had building the platform as well as, you know, the cutting edge, you know, techniques that have evolved over the past few months. So that is a program for someone who is, let's say a software engineer or you know, a data engineer who wants to get into AI engineering, they can take that program and emerge as a Kehli engineer able to build some of these LM enabled products. So that would probably be our focus for the next three to six months and then the next phase for us we'd probably be focusing. We're still kind of thinking through it, but most likely it will be focused on both AI literacy and data literacy. So here we're targeting much wider audience that is mostly non technical and the aim is to fill this gap. It's traditionally been a massive gap and we think the courses that were done before were, you know, just classical courses. So they were quite static and especially for these areas the variance is very, very wide. So if someone is, let's say a software engineer who wants to get into AI engineering, the variance is not that much. Like, the foundations are pretty similar and people tend to kind of know them in a fairly similar way. But for teaching, for example, data literacy, you have people who have never interacted with data before to people who have done some accounting. So they're familiar with working with sheets and with numbers, but not maybe with measurement and some of the statistical concepts, and then people who are familiar with those, but. But not to a sufficient degree that they can work with them effectively in their company. So we think that this new medium can allow us to deliver these courses around AI literacy and data literacy in a way that adapts more to the learner. And in this particular case, that can be pretty valuable. So at the moment, this is kind of where we're thinking about the content roadmap.

Okay, yeah, certainly I can see how SQL, Python are very important if you want to be data analyst or data scientist. And yeah, focusing on data manipulation would make sense because everyone has to do that. And then, of course, AI engineering, very hot topic. I'm sure there's going to be a lot of people interested in that. All right, thank you, Yusuf. And thank you, Jonathan. Great.

Thank you, Richie. Sounds great.

Thank you.