Creating AI Bedtime Stories for Children and Navigating EU Regulations - Matthias and Dima from OscarStories

On this episode we met with Matthias and Dima, the founders of Oscar Stories - a mobile app that lets parents create age-appropriate, personalised bedtime stories for children, made with AI. We spoke about how they managed to launch within just a few months of ChatGPT’s release and what it’s like to run an AI startup in the EU.

----------------------------------------
Welcome to WeAreDevelopers, the #1 developer community in Europe!
This is your one-stop destination for the latest tech insights, tutorials, and career advice to elevate your developer career.

Stay updated Dev Digest, with our weekly newsletter featuring the most recent tech trends, career guidance, and original content crafted by developers, for developers. Subscribe now

Interested in advancing your career? Browse through our job board featuring over 190,000 jobs. Unlock job opportunities with a free developer profile

Don't miss out on the annual highlight of every developer's calendar - the WeAreDevelopers World Congress. Network with 15,000 peers and learn from over 500 speakers. Secure your spot and save 10% with "wearedevs_yt"

#CareerInTech #Tech #ProgrammingTutorials #DevRel #CodingTools #TechJobs #TechTalks #DeveloperSkills...

Transcript

Welcome to another coffee with developers. Today we got Matthias and Dima here from Oscar Stories, and everybody and their dog and their neighbor is building an app with AI right now. But they've done it like three years ago and went through a lot of like, interesting things. The interesting bits here is that first of all, they're European company and secondly, their target audience is kids. So I'm not the one to talk about that because I'm handing over to Dan for that.
Because you sired offspring. I didn't. So it would be interesting to see if the product that they built would be something for you. So what are you building and how have you done it? Yeah.
So thanks of all, first for having us. Dan and Chris, thanks so much. Yeah, we have built Oscar Stories, which we basically released our first MVP end of February 2023. Well, the app basically already has 100,000 users. And what it does is you can create personalized bedtime stories for your kids where your kids together with their family, friends, pets, and whoever they want, can basically take a main role within the story.
The story can be read as well as created into an audiobook or print it out as well. So you just mentioned that our focus group or basically our target audience, our kids. That is true, but actually we have like our kind of audience or basically our target group emerged from that. So actually a lot of parents use the app as inspiration as well. I think I followed the journey of Oscar Stories really, since I think Matthias, I think we connected first and then I found da through you guys as a partnership.
Where did the idea come from? You have got something on your website about it coming from a family member, but you. I think that's an interesting little tidbit in itself is where did the idea for the app actually come from? So my godch child is named Oscar. I think I have a picture somewhere from him.
And this is actually where the name stems from, because Demon and me, we don't have children, but we are engaged godfathers of children. So Dema has Sammy, I got Oscar and. But we both realized that the respective parents of those kids were not really very creative and sometimes struggled with making up a bedtime story. So, for example, the father of Oscar is actually a state attorney, but he told me once, yeah, his child always wants to hear something new and has very specific ideas what the story should be. And he struggles.
And at that time, we were already experimenting with GPT3 and working actually on a different project. And ye. And then Dima basically said, hey, we could actually try to make baddime stories. That's interesting. I mean, you're saying about the model as well, that was going to be one of my first questions is you said about 2023 was when the MVP was launched, but it got me thinking about how long before was it that you were actually experimenting with this?
Because I think for most of us it now feels that we're at a stage whereas users of AI, just front end users logging into chat cptt, things like that, we understand how it works, we've all got a good handle on it. But I can possibly speak for other developers in saying that some of us aren't quite ready to start building things with it because it feels like the ecosystem, the documentation, know the tutorials that take you all the way through it aren't out there in as much quantity as they are for other things. So how did you then jump on it and start using it? Did you go straight into using the API or did you sort of try and experiment with, with the UI and see what you could do? Ye, well, at first I think everyone remembers how ChatPTT launched in December, I think it was December 22nd if I remember correctly, or something like that, or maybe November 22nd.
So shortly before that we already had GPT like generative pre trained transformers on our radar and we'experimenting with them. But once ChatPTT launched and once it actually, you know, was built to chat with you or to talk with you, that's when we started experimenting and started building. I think it was basically end of December 2022. Directly with the API, we built an MVP in the React native for iOS and Android. We both knew it has to be an appet that can't be a web app, it has to be a mobile app.
So basically we very quickly, within a week or two we just made an interface with, you know, enter a name button, get the story and then started experimenting from them. And Mathias can tell you that our backend already is much more sophisticated than just an API call. But yeah, you're right. Back then tutorials were very sparse on that topic. That's true.
I mean an API is an API, of course, for a developer, but still getting it right or getting the quality of stories to a very high level because kids are the worst critics. Yeah. If they don't like something, there is no second chance. That's it, like you're done. Yeah, I remember getting early user feedback when you would get, I don't know, the gender, not right in some part of the story, which GPT3, especially at that time was really Struggling with.
They were like instantly saying, yeah, this is not for me is when one sentence was not perfect, you were out, you're done, basically. So that was a lot of work. And yeah, our first MVP was really, really basic. But we knew we had to go out fast because as you probably know, we are not alone in that space anymore. There are a lot of people trying to do similar stuff.
And we were also pretty lucky to be fast and also get some media echo and grow, which probably wouldn't have worked if we would have taken longer to get feedback to UX UI reviews. So we were basically launching the app even without payment at first. Just be on the market and get feedback. Why did you say you have to be a mobile app? Is it to be available on the marketplaces or is it just a growth thing?
Or are you worried that basically when you're on the web you get too much traffic that you can't control? It was more actually from a user experience point of view. I mean just, I don't know, sitting somewhere with the child. Like our first idea was like, yeah, the parents would sit together with children and be like, hey Sammy, who do you want to be? Hey Lisa, who do you want to be this time?
And then they create stories together and that's just much more convenient on a phone than a web app. Also, as you mentioned before, Daniel, it's also about having a kind of responsible experience with the children. This is why we also have the audio stories part, so people can connect the stories to their, I don't know, Bluetooth device. And the child doesn't necessarily have to see the app. I mean we have all the children use the app themselves.
That's why we have also parental controls and you can also print out the stories. So this was one of our earliest features that users wanted to. Because seeing the screens and children is a very sensitive topic, of course. And so we are mindful about that. That I mean it's kind of counter intuitive as an app developer.
But we don't want our users to spend too much time on the app but use the time the app regularly and have a fun experience but at the same time not have children be stuck on the screen for too long, basically. I mean this also hard to find balance, I would say. What languages do you support? Do you have more German than English users or French, Italian, Hungarian or things that don't work at all? And how did you find that ChatGPT can deal with other languages?
Yeah, so currently we started off with English only, then we added German and then at some point. So basically, of course we got a lot of press once we launched and therefore, obviously Germany or Austria in that case, Austria was our biggest market at first, then followed by Germany and then the US took over pretty quickly because they are, you know, they're much more open to things like that. So the up grew there much faster. At some point we started adding other languages such as Spanish, Portuguese, Italian, French, so basically the Western languages. And currently Portugal, Brazil with Brazilian Portuguese is our largest market.
So we have by far the most users there and we've never done any ads or anything like that. And now our plan is within the next few months. So we're going to Hong Kong with the Austrian Chamber of Commerce. I think they're basically invited us to go to Hong Kong and we will be expanding to Asia. So we have added also some Asian languages as well.
That's so fascinating. There's so much to unpack in there. I mean, I'm also personally quite happy to hear. I can see Chris is smiling. You know, my daughter is going to be.
She's very young, so she doesn't speak any language yet, but she's going to going to be speaking Brazilian, Portuguese and English. So before doing this, my wife said, I told her about the project and she was like, yeah, that sounds great. Can you check that they do Portuguese? So you've answered that question. The.
I mean, I just want to go back a few steps though, because I think it was so interesting some of the stuff you were saying about the. That when you were using those models and you wanted to go to market quickly and you needed to get that feedback from parents, but you had to balance those two things of you didn't want to miss the boat. You know, how easy was that? Because I can imagine there was quite a lot of red tape. But also just actually getting control of the LLM so that it's producing those, you know, those good responses and that it's not giving you things that are inappropriate for children.
How did you balance that, that SPE lead and responsibility? So one thing we do differently than most apps or products in the market is we don't allow completely free input or completely free prompts, which actually has proven to be the right choice, even though some requested. Because we have an input control for the free text fields like before sending them to the lm. And we also have some output controls, so we checked the output before sending it to the user and which was implemented pretty early. I think we have a feedback system for every story which users actually use.
I mean, the only benefit they get is for us, but in the long run for them as well. But we can really track if certain. Right now we have kind of an AI agent system to generate the story. So we have some, we call this storybloc. So it just generates an introduction, a middle part and then an end part.
And then also checks for character consistency, which is super important. And what we basically see when we add some new features, we also have some different kind of universes. We can really track other problems to people really like these stories. Are the new stories consistent on scale? Because of course you just tested for, I don't know, 50 stories, but it could happen at the 51st story, something goes wrong.
So I think for LMS, basically keeping human in the loop is super important. How do you do that and how does it scale? Like do you have a team of editors that look through that stuff or do you have just a pattern matching algorithm thats if there's naughty words in there. So we check with an LLM if there are naughty words inside the text. But I mean most biggerms have basic guardrails.
And what we are right now working on is we, we're working on our own fine tuning of an open source model which would be LA 3.2 or 3.3 and focusing on child appropriate language. And we have a collaboration with an education institution in Germany calledadem and they're basically just giving us free feedback and they're super nice. So shout out to. I put their very long domain name somewhere. I'm sure it's, it's very German, very long mo, five lines like, you know, gu.
I just want to move on to the talking about some of the things that are actually in the footer of the Oscar Stories website, which is some of the kind of specific stories that you guide people towards creating. So there's ones like teaching children about climate change stories for teaching children morals, reducing anxiety. Are these things that you found are important through, you know, feedback with the users or are they coming more from a personal value perspective? What was your reason for wanting to kind of bring those, you know, into the forefront? Actually, I would say it's a combination of both Mathias and I.
I mean everyone, I mean it's understandable that Oscar Stories, the way it is right now is let's say 70% an entertainment product and 30% an educational product. I would say yeah, I mean we tryied to build like modes for, you know, various things like teaching children about vegetables and stuff like that, as you mentioned before. But we always wanted to Go more into the educational space. And that's how the idea for our new app, which is basically how to say, in addition to Oscar, which will be called Laura, where we will be teaching children STEM subjects through storytelling. And therefore we will basically, as Matthias just previously said, fine tune an open source LLM.
Because child appropriate language is very, very important in that case. Because if the child at 5 years old or 6 years old reads a word like antagonist or something, and especially if we're trying to teach it physics, then there shouldn't be any kind of difficult words in those texts so the child can actually concentrate on the story. And the idea here is to teach children those STEM subjects through storytelling. For example, such as teaching the concept of gravity by Isaac Newton within the setting of the child's favorite current setting. Like it could be, I don't know, a forest or it could be a construction site or whatever the child is into at the moment.
Yeah, it's a combination, of course, because we know parentsnn want their kids to be smart. Right? So that's number one and number two is Matthias. And I just do it as a personal, you know, thing. There's a few great products already out there, like have you heard of hello Ruby by Linda Liuckas?
It's basically kids books that teach you about programming and also teach you about other things. And it has Ruby as the main character and she's like this unruly girl that has actually started in Finnish and now it's in English as well. It's really good. And I mean, when you work with the official government things, I mean, TRD we got Logo, which is like the news for kids. And we also got like Kurzgesagt, which is like in a nutshell, the five minute videos that are animated.
So there's lots of stuff that you can work with now that you've got the foot in the door with the government. Would be interesting to see what else you have. I don't know if you have similar things in Austrian, but it's simplified German, so you should have those as well. So I know some of them. And I mean, it's not.
Of course, some of these collaborations take a long time, which is also hard for us as a startup because the timeline of official institutions is different than what you have as a startup, of course. But yeah, we are trying to also find more collaborators and also get more data for fine tuning, because getting high quality data for that is also not that easy, I have to say. I mean, there is some public domain data out there, but you Wouldn't want to fine tune on the original Grimmms s tales because I don't know if you've read the original ever. They are pretty gore for children, I gotta say. They're more like moral stories than kids stories.
They're just basically pushing people in a direction. It's just really bizarre how these things. Yeah, it is, it is. You know, we hear so much about startups and there's so much conversation at the moment, particularly around startups and regulation and particularly in relation to ones in Europe. And so much of the conversation we see this, I'll say primarily on X.
Okay, that's pushed very heavily on there, that narrative. But we do see a lot of the discussion saying that you entrepreneurs in Europe simply are so bound by red tape that they can't do anything innovative, they can't move fast enough to catch up and the regulation kind of binds them to effectively just giving up. You're both quite outspoken really about not pro eu, but I think trying to maybe give some other side of of the conversation. How have you found it actually being founders and trying to get something off the ground in the eu? Is it as bad as X wants me to believe?
Very briefly, from my point of view, the things that we are doing. I mean, yes, of course, maybe you have some sort of, you know, reporting things where you'd have to basically report on what kind of data you've trained your model. But I mean, unless you're actually building an LLM from the ground up, which I think none of the indie hackers or none of the people on X are really doing, I really don't see any kind of specific tape that the EU puts around you in that case. I mean, I'm not even close of to building my own LLM. Yeah, unfortunately, I mean, because that thing is of course we know the new models steep seak and so and so on made it cheaper, but apparently cheaper, but it's super expensive and that's a completely different beast right there.
Yeah, I have to say I think there are some, I would say uncertainties, for example about the EUEI act that could pose problematic for innovators. But on the other hand, I see a lot of discussion revolves around how bad is the EU and how bad regulation is. But in the end, regulation for the eu, it's very important that the EU regulates and harmonizes because most of what people are talking about and say this is hindernce and this is a problem for me it's actually national regulation and that hinders us a lot. I mean, there are cool initiatives for example for havingified one unified type of cooperation in whole Europe. Because that is a problem of course we different investors, you have investors from overseas that don't understand the legal structures in your country and it would be great to actually have more regulation in that sense that you don't have a different kind of rule set in Germany to Austria to I don't know, Spain.
So I think we should actually focus on having more regulation but more unified and haronized regulation. And of course there's always a trade off. I mean the European Union is very customer focused which is good from a perspective at as a citizen. And it can happen that some of those rules would affect businesses. But on the other hand, and I think that always gets lost in these whole discussions in almost all other indices like I don't know, freedom of press, quality of life, criminal statistics, you countries are in the top, top 20 or top 10 and you have to find a balance between having good growth and there are problems in Europe there.
I think Europe as a continent has missed some innovations or some new technologies. But that doesn't mean we should just give up on everything else we have achieved and giving up on the quality of life, giving up on health care and all that stuff. I mean I have lot of friends from the US who are right now really, really concerned about what is happening. And yeah, I don't think the average person will see a lot of the increase in stock prices from Nvidia for example. And yeah, so that's suggest to put it in nut sh overview.
It's an interesting point though because the hyper growth that actually is necessary in America and especially when you're in AI startup and you go to VCs you actually compared to the American ones that can move much faster or the Chinese ones that can move even faster because they don't have to care about things like workers rights or like these kind of things. That's the thing that is always a comparison that's unfair. I think that's why there's a few European VC funds as well. Thinking about like regulating that and going for it. Did you see that you actually are being.
I don't know how you're set up in terms of as a company but do you see that you're being pushed to move much faster than you feel that make sense for the product or is it still okay for you at the moment? I mean we don't have any investors. Yeah, I mean we were, we thought we need investors but at some point we were on a TV show With Oscar stories, which is basically like the Shark Tank. Like Shark Tank, something similar. So that's when the growth took off from us and we kind of decided, all right, yeah, it's easier for us to stay bootstrapped.
But it is true that once you start looking for investors as a founder and if you decide to go that route and I think that decision really has to be thought out and not just like a snap thing like yeah, I'm just going toa go get investors because that changes your life significantly. You then not only have to report to someone, but then you also have to adjust your growth goals to satisfy other people because maybe those other people aren't investing their own money. If you have angel investors, that's obviously a little bit different because those are just rich people more or less investing their own money. But once you'grow and grow and grow, then obviously they would push you towards an exit because they have to know, make a return on their investment. But for bootstrapped founders, I mean you grow as much, as much as you want to spend at the end of the day.
Yeah. I also have to add one thing here and that is something that is really good. In Austria, for example, we got pretty early state grants which are non refundable. Without them we would have had to get investment. And that was pretty amazing I have to say think sorry, was that the federal industry for digital?
Because I saw that on the website and I was yeah, exactly. And I think that is really one of the advantages of having a company here. I mean there is one downside I have to say, which I see from or I hear from a lot of my friends that the timerame for getting investment in the US is way shorter than average then in Europe. So you have longer talks, you have much more, you have to report more. And I think most people talk about that the regulation is a hindernce, but I think mindset is the bigger one or the more important one.
So I think in the EU we are more risk averse also in the investment scene. And I think that is something if that changes or if people would invest more in their own companies. So most of my friends, they invest in NASDAQ but not in EU stocks and then complain about the lack of growth. And I'm like, yeah, but dude, if nobody invests, how should they grow? This doesn't work.
Yeah, I mean the whole idea of like, I mean I invest in like renewable energies which is not moving fast up and down, but it's moving steadily and I think that's an important thing. To do the risk adversement, I think, is an interesting one. The other thing is also, like in America, if you're ceior entrepreneur and you basically ren four countries, four companies into the ground, people still keep believing that this time you will actually not make the same mistakes again. Whereas, like, you make it for the first time in Germany and you're basically done. There's no way you can get money anymore.
So having a clean exit is a different story. But having brought basically a startup not working out, it means you have to go out there. And it's really frustrating because I see it on LinkedIn as well. A lot of people are just like right now going bonkers about, like, how America is so much faster than Europe is. And then I'm like, so you live in Hamburg.
Why do you live here? Just go away if you like it that much and you go over there. So, I mean, Germans are unhappy when they can't complain. Like, that's the main thing. And it's like we love to complain and compare these things all the time, but we don't do much about it.
So it's great to see. It's interesting that you were on Shark Tank or Dragons Stand or Lion thing. It was TR in Europe. I don't know what are different animals, but there's oftentimes things that are focused around kids there. I remember those, those trunky like suitcases for kids.
They were there as well. In Germany, not in England. Back then, the Dragons Stand, as it's called there. How did you feel when you've got investment there? Did you feel like you had to act fast?
Were they pushing to get the thing out soon or is it more of a. You had more to do more interviews than coding afterwards. We actually. We didn't get investment there. Yeah, we.
And we all. We. How to say we almost knew that we're not going to get investment there because it's called two. Two minutes, two million. Basically, the show in Austria, which is very similar.
It's basically the same thing as Shark Tank and the investors there, they weren't focused on investing in apps. Yeah. But it was still a very huge PR success for us. So we didn't get invested. We didn't get the money.
But that was perfectly fine for us. We got the exposure. And I just remember, like, as soon as they were showing it live, the downloads and like the notifications of people buying the app just went crazy. The moment when the show stopped and they did like advertisement break for like five minutes, the graph was just like up. Yeah, I mean, I have to say you can say a lot about things, about, you know, old school advertising and stuff, but TV just kills everything else.
I mean, we were featured in Wired, like in one article and got a backlink from there and it was not even comparable to the amount of new users we got from the Austrian TV show. It's crazy. I mean the impact that has on people to see someone on TV is just also, I think some els to with trust I guess because they're like, ah, you are in tv, you're serious. But I actually wouldn't have thought how much impact it had. I guess it's a local connection as well.
They see like two guys with Viennese acc and talking on taly and they're like, oh, that's an app built here, so I'm going to use that one. And they got the phone in their hand and it's one click to basically get it. So that might help as well. Yeah, and we get that feeling more and more. I mean, yes, on Twitter and on X there is this anti EU feeling in general.
But for example, we have another app which is called as PDF, which is basically a PDF chat and like an AI assistant for students which we did user testing for in Germany just to about two weeks ago. And I'spoken to 10 people and actually half of them were like, oh, that's an app made in Europe, that's an app made in Vienna. Oh, I'm definitely going to try it out. And they were actually very, very happy, very excited about ##haps, being built here. And I think that, or I think, I hope that the mentality here basically kind of start shifting towards startups or towards supporting startups from the European Union or just Europe as a continent in general.
We need more success stories. I mean Spotify was always the big one, like o Swedish company, but it's been an American company for so long now. There's not much left of the Swedish original idea, I think in there because I, I knew the founder back then and he said like I want to make illegal downloads of music unnecessary. That was the first goal of Spotify. And now it's like which podcast can we have and can you play games on?
I don't even know what's what all the features are by now. So do you know if they're profitable? Not, I don't know. I mean I would be, it would be very surprised if not. But then again Amazon wasn't profitable for like 15 years.
So it's. You never know, you know. And I think in this case is also the biggest problem with music distribution, film distributions, that you actually have to pay the studios for the right to do it. So I guess that's where the money goes. In your case, it's like user generated or LLM generated content.
Which brings me to another question. When you said earlier that we're actually testing these things and running it through an L&M, do you also do collect stories? Is there like an idea to say like, hey, here's a success story that what kids liked a lot. Like you say you start with blocks, but do you also think about like making it reusable and saying like here stories. If you just want to listen to one rather than create one, that might be a good opportunity to reuse the content rather than having to recreate it every single time.
I mean, we have thought about it and I think it was at some point in our pitch tech as future ideas, but we never ever actually got to implement something like this. I mean, we were thinking about having children, book authors distributing their own stories on our platform or something like that. But right now we want to focus more on creating Laura and more on the education side because we think that there's more value in it than just focusing on entertainment. But we will do both, of course. So that is right now the idea.
But yeah, I think it would be great to do that. Of course it would be a cool idea also to combine LM generated content with human written content and just having kind of a story marketplace. Yeah, lots of ideas. I ll copyright that one. Chris.
Just to finish up then. Just to finish up. Chris Y I'm a Creative Commons person. I don't do copyright. I'm sorry, I like.
So just to finish up, what does. It's a big question really, but what does 2025 look like? Ve you've given us some idea of kind of what projects you're going to be focusing on. But what's the big goal then for this year? One of the big goals is finishing up Laura.
Launching Laura, that's going to be, I would say, Q3 of this year. Once we're done fine tuning, once we've done the app, once we did some user testing, we're obviously going to leverage. For now, the plan is to launch it as a separate app and we're going to try to leverage the user base that we already have at Oscar. In general, I would say, if in very brief, to steer towards more educational content and more educational space in 2025. Yeah.
And we also got a slogan for this year which is more collaboration. So we were actually planning on talking more to educational institutions, other startups, finding some novel ideas, because we also realized it makes no sense just to try to make it on your own. And why not find other people who have in similar space and have some fun ideas together? Yeah, and also we are going to Asia, as we said previously, so we'll see how that goes. And yeah, it will be interesting to see what kind of, what kind of ideal stories, what the difference of the, of the stories that are there.
I would love to see like from, from a linguistic point of view, like what people create in Asia compared to Europe. There's a super interesting story in there in terms of technicalities. I mean, we have a technical conference coming up as well.
Do you have any idea how you can share stories behind how you scale the thing, how you build the thing, what developers can learn from your story if they want to get something like that started? Da and me did a talk about bers in AI and how. And child appropriate language and how we kind of focused that in terms of bit technical and also in terms of the general idea with Laura. All right, well, thank you very much for joining us today, Matthias and Da. Um, good luck with Oscar stories and yeah, I'm sure we'll see you again soon.
Yeah, thank you so much. Thank you. Thank you. By.