COBY’s Child-Safe AI Architecture for Bedtime Routines | Startuprad.io

What Is This About?

COBY has built a child-safe AI architecture specifically designed for bedtime routines — creating interactive storytelling experiences that parents can trust. The startup addresses a critical gap in the AI market: building products for children that are genuinely safe by design, not just filtered after the fact.

Introduction

Building AI products for children demands a fundamentally different approach to safety, privacy, and content moderation than adult-facing applications. COBY has developed a child-safe AI architecture specifically designed for bedtime routines, creating interactive storytelling experiences that parents can trust. This interview explores the technical and ethical decisions behind building AI that entertains young children while maintaining rigorous safety standards that satisfy both regulators and parents.

COBY built a child-safe AI architecture specifically for bedtime storytelling that meets rigorous parental trust and regulatory safety standards. The system generates personalized interactive stories while maintaining content guardrails that prevent inappropriate outputs. Technical safeguards include multi-layer content filtering, age-appropriate vocabulary constraints, and parental oversight controls. The product addresses the growing market for AI-powered children's content where safety is the primary purchase criterion.

COBY shows how to build EU-compliant child voice AI that earns trust before coping, reduces bedtime conflict, and stays privacy-first by design.

COBY shows how to build EU-compliant child voice AI that earns trust before coping, reduces bedtime conflict, and stays privacy-first by design. Startuprad.io brings you independent coverage of the key developments shaping the startup and venture capital landscape across Germany, Austria, and Switzerland.

This founder interview is part of our ongoing coverage of Scaleup Founder Interviews from Germany, Austria, and Switzerland.

In this coversation with David Cañadas Link, we discuss Child-facing conversational AI only works if safety, privacy, and trust are enforced as system architecture and validated before release. COBY’s model is explicit: earn trust through positive interaction first, keep data control in the EU, and measure success in household outcomes like time-to-bed and conflict frequency.

• Trust must be built before a child-facing system asks about “problems.”

• Privacy-first child AI is an architecture decision: compute placement, filtering, and EU hosting.

• Screen-based coping tools often fail because they produce passive consumption, not regulation.

What makes child-safe voice AI defensible

Defensible child-safe voice AI requires pre-release proof that the system resists harmful dialogue across adversarial prompts and real-world child triggers. COBY enforces a dual gate: agentic simulations that attempt unsafe conversations at scale, plus expert validation before any release.

Agentic red-teaming stress-tests the model across thousands of attack patterns, including sexual and harmful topics, to verify refusal behavior is stable under pressure. Expert validation then tests qualitative edge cases drawn from clinical experience, catching failure modes that synthetic prompts miss.

David Cañadas Link describes child safety as a system of gates, not an aspiration, pairing automated adversarial testing with expert network review.

Why trust sequencing matters more than “coping features”

A coping system fails when it asks for vulnerability before trust exists. COBY treats trust as a staged interaction: positive engagement first, problem discussion later, because children reject direct probing the same way adults do.

Trust sequencing is a conversational UX constraint: early interactions must create safety, predictability, and positive reinforcement. Only after repeated non-threatening experiences can the system introduce reflective prompts about feelings or problems without triggering shutdown or avoidance.

David Cañadas Link describes direct “tell me your problems” prompts as a predictable failure mode and designs COBY to earn disclosure gradually.

Privacy-first AI is compute placement, not a slogan

Privacy-first child AI depends on where data is processed and who controls the models. COBY avoids uncontrolled external hosting, uses EU-hosted AI under operator control, and shifts speech processing and filtering to the parent phone to keep device cost viable.

On-device AI can force consumer pricing to levels that break adoption. Server-hosted models raise compliance stakes and data governance requirements. COBY’s compromise is layered compute: minimal device functions, parent-phone background processing for STT/TTS and filtering, and EU-hosted inference for controlled model behavior.

David Cañadas Link frames privacy and compliance as architecture decisions that also determine margin, battery constraints, and latency.

Screen-free is a behavioral constraint that changes outcomes

Screen-free devices can reduce overstimulation and prevent passive “zombie mode” because interaction is tactile and dialog-driven. COBY’s plush format supports bedtime use and encourages two-way participation rather than one-way content intake.

Screens are optimized for attention capture, not regulation. A plush companion can be held, trusted, and integrated into rituals, shifting the experience from consumption to interaction. Two-way dialogue forces active participation and makes routines coachable rather than distractive.

COBY’s choice of a plush companion is justified by child testing: kids engaged more and felt safer compared with screen-based or “robot eye” prototypes.

What you measure predicts whether the device becomes a routine

If a child wellbeing device does not change household bottlenecks, it will not retain. COBY uses routine metrics—time-to-bed, evening conflicts, and homework minutes—because they reflect real parental load reduction and child regulation improvements.

Engagement metrics can mask failure in the only place that matters: the household. Routine metrics expose whether the system reduces friction at predictable conflict points. They also provide decision-grade evidence for whether the product is a “nice idea” or a durable habit.

David Cañadas Link cites time-to-bed, evening conflicts, and homework minutes as core outcome metrics tied to retention.

Inline Micro-Definitions

• Emotional paraphrasing is the act of naming and reflecting a child’s emotion back to them to validate and reduce escalation.

• Agentic red-teaming is automated adversarial testing that tries to force unsafe outputs across many attack patterns before release.

• Phone-as-compute is a system design where a smartphone performs heavy processing for a peripheral device to reduce cost and latency.

• Safety gate is a release constraint that blocks deployment unless defined adversarial and expert validations pass.

  
  

Operator Heuristics

• Do not ask for vulnerability before the system earns trust.

• Treat child safety as release gates, not policy text.

• If privacy raises cost, redesign compute layers before raising price.

• Measure household outcomes, not engagement minutes.

• Make parents the authority layer by default.

• If latency breaks trust, move computation immediately.

WHAT WE’RE NOT COVERING

We are not covering general screen-time research, general parenting advice, or generic “AI for good” narratives because none of these determine whether a child-facing conversational system is safe, compliant, and reliable at bedtime. We also do not cover broad toy market analysis because this category is governed by trust architecture, not entertainment competition.

Relationship Map

• Jörn "Joe" Menninger → Host of → Startuprad.io

Automated Transcript

1 Child doesn't need more content, a child needs a safe 2 mirror that turns chaos into calm. 3 Coby is building Emotional Tech for Kids, an 4 AI powered screen free companion that 5 helps kids ages 4 to 12 6 navigate emotions and routines without 7 overstimulation. The conflict is simple, parents 8 are exhausted, children are dysregulated and screens 9 escalate the problem. David Kanna's links by 10 Bild Kobe from Lived experience and engineered it 11 as a privacy first EU hosted and clinically 12 informed positioned for dach early adopters and 13 TikTok native go to market. The payoff is a new 14 category with real unit economics and not a feel 15 good gadget. Foreign. 16 Your podcast and YouTube blog covering the German 17 startup scene with news interviews and 18 live events. 19 Today's guest is David Cañadas Link Is 20 that right? Candidas? That's right, that's my Spanish

21 surname. David Cañadas Link | CoFounder & 22 CFO of COBY - The Coping Buddy, Coby is not entertainment, not therapy and 23 not a toy. It's emotional technology screen free 24 child tailored AI inside a physical 25 companion with parent control and boundaries designed 26 into your system. This conversation sits 27 at the intersection of voice, AI, child well 28 being and consumer hardware where trust, privacy and safety 29 are not marketing points but product architecture. 30 David's true story begins with a progressive diagnosis 31 and a hard question how you stay 32 grounded when certainty disappears. That 33 personal inflection became a product thesis. Children 34 live inside big emotions daily 35 and most households lack twos to 36 to de escalated state of distract. Cobie's approach 37 is clinically informed dialogue, predictable routines 38 and gentle ritualized close of the 39 day delivered through plush companion that avoids the screen

40 war entirely. We cover Category creation Privacy 41 First AI design under EU Expectations TikTok 42 as a distribution engine and what 43 it takes to ship hardware with real margins while staying ethically 44 clean around children. 45 That was already a lot. David, welcome to the interview. 46 Thank you so much for having me. What event 47 forced the first irreversible shift 48 in how you think about uncertainty and control? 49 Amazing question and yeah, there's 50 quite a personal and life altering story behind 51 that. The event it was five years ago, I 52 was 29, I was wanting to make my 53 fur shine so my driver's license for car I had for 54 motorbike and yeah, and I realized pretty fast 55 that the person sitting next next to me was seeing way more 56 than me. Like it was hinting me to why didn't you see that sign

57 and that sign and there was a person I was like I don't know. Like 58 I don't know what you are telling me. And yeah, it went pretty fast. 59 One Thing led to the other. The eye doct didn't want to tell me 60 anything, sent me directly to university clinic. 61 They conducted many tests and yeah, suddenly I 62 was sitting in front of the professor. He told me I have the diagnosis Mobile 63 Stargardt, degenerative genetic 64 incurable eye disease. That I was gonna lose 65 more sight, central vision, but I shouldn't 66 be too afraid because before I go totally 67 blind, I will already be dead. That was a fun 68 conversation. That was a heavy conversation, I have to say. 69 I see, yeah, more of you were expecting to get new glasses, 70 right? And yeah, I mean, from there my personal

71 journey started. And I mean, the life 72 changes completely, right? You ask yourself what I'm gonna do with 73 outside. I also like to paint canvas 74 and pictures like how, how am I gonna do this? Or how am I gonna 75 work? Will I be seeing the faces? Right. 76 And the heavy part is that really once that at least in Germany, 77 the medical system can't apply any more therapy to you. 78 You're completely on your own. So there's no one helping you 79 to adapt to that condition. How are you gonna 80 live with it? How I'm gonna read, how am I gonna 81 be mobile, like mobility and also how 82 I'm gonna deal with my emotions regarding the disease and the 83 progression of it. So. So basically how I'm gonna 84 cope with it. I have to say there was a

85 long journey from there of talking with many patients being into 86 patient communities like the PR retina, that really helped a lot. But 87 where I also realized many people that are there 88 are elderly, don't even use computers or 89 smartphones. And it's really hard to 90 really match with someone that actually understands. Even though if you have family and 91 friends, no one has the same experience as you have. No one 92 sees what you can see, no one goes through that. 93 Yeah. Also the experience of losing sight and 94 it's a very, very difficult road to go along. 95 In my case, I think I was quite lucky in the sense that I'm 96 quite a tech person. So I found my ways of adapting, 97 right. Of finding the right tools on my smartphone, on my laptop, 98 I'm right now zooming in and out all the time with my gaming mouse to

99 see your face and so on. But still, 100 I noticed there's a big lack in this right now. From 101 there on, many things happened. I talked with many patients, not only 102 individually impaired community, but also through consulting projects in other 103 communities from back pain over asthma, over 104 diabetes. And there came a point where we 105 realized as an adult at least you have, let's say, 106 the resources or knowledge to find for yourself solutions or get 107 in contact with people. But for example, with asthma kids, and that was 108 when Kobe really was born, we noticed kids don't have that 109 type of resources. So here the problem is way bigger. 110 Like if you have a heavy diagnosis or chronic condition as a 111 kid, you're completely dependent on parents that want to help, 112 but often don't have the time or knowledge to do so.

113 Physicians or experts that you see all two years, if you are lucky 114 and that are really quite technical with the input that you give, give 115 to you. And then your peers being other kids, 116 usually you are one of a kind. So it's extremely difficult 117 to find someone to relate. So we decided that's something where we really 118 want to find a solution, want to make a change and 119 give this kids a body, a companion on 120 eyesight that can help them through this journey. 121 I see. And we may add, if you do have sick children, 122 it is that the parents are more busy in providing 123 care for them than really enable them 124 to meet other children with the same condition. That's really a 125 difficult point. I was wondering, 126 are you willing or able to share because the

127 doctor said you 128 may not live to completely lose your 129 eyesight. What time horizon are 130 you looking at here? I think in my case 131 I have to say now looking back 132 that I think I'm still quite lucky in a sense. Like with the live, 133 it just meant in a sense that I will live long. It doesn't affect 134 now my normal lifespan, but 135 the progression will be slow enough that I will not lose 136 my total eyesight. So visual impairment is a spectrum 137 like in most diseases, and in my case, usually you have 138 visceral impairment. Either you have central vision loss, what I have, 139 so it's mostly known as macular degeneration, or you have 140 peripheral loss. So that's mostly known. Retinitis 141 pigmentosa. When you have periphery, then it's really about 142 mobility because you cannot see any more objects in your, in

143 your periphery. So you don't see a stone on a chair if it's down 144 beneath you. So you will need a white cane. Or for example, you don't 145 see a car coming from left that. Exactly. 146 Now I see cars coming from the left, but I don't see details. So well, 147 in my case I cannot read. So the progression was pretty 148 fast. At that time I had, when I was diagnosed, I had 149 80% of visual acuity on my left eye and 60 on the right 150 eye. Now I have 10 on the left eye and five on the right 151 eye. So I'm officially impaired in Germany with 152 70%. But I am very well adapted. 153 So the good thing is I use many technical tools 154 to zoom in, zoom out and also in my disease, most 155 of the times that's kind of the main

156 like the main problem is the central vision loss because it doesn't expand 157 too much over the whole retina. So hopefully that's the case. You never 158 know for sure. But yeah, 159 I hope it will not get much worse than it is right now. 160 Yeah, fingers crossed for you. 161 Going a little bit back to the story of Kobe. What 162 problem definition made you decide that children 163 need like a physical companion and not an act? 164 Ah, that's a good question. 165 So if you want to build a coping companion, 166 meaning someone that you can also talk about problems, it's someone you will 167 have to trust. And children to 168 build trust they need something real like 169 screenshots is something that 170 actually nowadays I have myself a little daughter. 171 Parents really in many cases it's the last resort they have

172 for the kids, but they really want to avoid. And it's also more like 173 a dopamine enhancing like entertainment device, 174 distracting device, but it's not like a real companion or 175 something you can identify with now also from 176 therapeutic and psychological inputs that we have with the experts we 177 were working with, puppet therapy is something that's extremely 178 effective and it's also used a lot in clinics. And the reason is 179 these toys, puppets, small toys for kids, 180 they are on eyesight, they open way more to these 181 toys that even to their parents or to the physicians 182 because they feel more like we are on the same 183 hierarchy level. Right. In many therapy, the many 184 therapists explain to us that in the therapy sessions they have 185 the first half hour is just bringing down the child to

186 really open himself because it's so difficult. Right. They are 187 scared. What does this person want from me? And so on. So 188 really finding a right medium in the sense of cuddly 189 plush toy that they know from their normal child life 190 was one of the. Yeah. The key points where we also saw 191 in our testings that kids engage most. I mean we had different 192 prototypes. We also had hardware prototypes with 193 amazing eyes and screens. And the eyes were interactive. 194 Adults loved it. Kids were scared. Kids 195 want something cuddly they can. 196 Yeah. Relate to and that speaks their language. 197 Yeah, the relation is very important. 198 So I was wondering when turning a product everybody 199 would, everybody else would see as a steady phone call, as an 200 app or something into a screen free product. What was

201 the first high stake decision you have to make 202 to narrow Cobie into this screen free product 203 and not like any Other screen based experience. 204 I mean the thing is we, we also made, made our market research. 205 I think the main thing was really user interaction. So conducting 206 many user tests now my co founder Philip, 207 picking up kids from the park, their parents and we're 208 showing them different interfaces, right? We showed them for example 209 a tablet with a talking toy or plush toy and so 210 on. The hardware robot and kids were always engaged in the 211 plushie. So that was one of the things. Also looking into the market, 212 what other products out there were also are very 213 successful. You see things like the Tony box, right? Also 214 completely screen free, telling stories or the tip toy, very

215 successful product here in Germany 216 that worked extremely well and still are working very well. And one 217 of the main points is really having the screen free. 218 The screen free component, it's something that parents 219 love, but also kids, for them it feels way more natural 220 actually to interact with screen free devices. If you 221 think about it, I mean as an adult we're so accustomed to these devices. 222 But as a kid you come with your 223 hands, with your eyes, right? With sensory, 224 other sensory like attributes to this world that are 225 actually not that fitted for 226 screens. They are more fitted for something I can cuddle, I can 227 take into my bed, I can take with me, like I can look at them, 228 I can have a conversation and it feels more natural. 229 So yeah, I mean we learned most from our customer from really the

230 kids were the ones that took the choice of the 231 plushie and still are doing so. 232 And also the market seeing that these products are not many 233 but are really something that parents are really also looking 234 a lot for. I know from my daughter, she's now two and a half 235 and how I feel like I've done it really only seldom times, 236 but two times I had my smartphone and she looked, I don't know, 237 Pew Pew was 238 some song. And I really like, she's, she's 239 instant and zombie mode. And I feel so bad if I see that. 240 Like as a parent you see that and you're like, I don't want this for 241 my child. I want her to, to be happy. And 242 also the next point, maybe that's not directly a point for the 243

plush toy, but a really important point for the AI and the interaction 244 itself is that's a two way interaction. 245 It's not one way, it's not you're just getting input 246 and you're just digesting a story or a song or whatever and 247 you're like okay, that's it. It's something where 248 the eye always gets into interaction, makes short 249 sentences, it paraphrases the kid but it always asks the 250 kid to follow up on the conversation. So that makes also a hu. 251 Huge difference. When. When. When you watch this 252 because you're noticing my kid is not in zombie mode. My kid is 253 having a conversation, creating a story with. With someone 254 by itself. And that also makes a huge difference. I 255 see. And not having zombies is always pretty 256 good. There 257 is one sentence you thought the model to

258 never say to child and it changes everything. 259 Editor, cut this out. There was a world teaser and 260 apparently that one was bad. Forget it. David. 261 Let's go to the next question. 262 What moment in early testing exposed the gap 263 between what parents say they want and what actually works at 264 7:30pm? Ah, you 265 mean with our entry use case 266 going to sleep. 267 I mean I have to say many parents know that what they expect 268 from the nighttime is very different from reality. Already 269 we notice a lot like going to sleep. Bringing kids to sleep 270 is one of the most crucial moments in 271 everyday's life for every parent you expect fact 272 that everything goes by itself, right? But it's not only going to 273 bed like it's brushing your teeth, it's putting on the pajama,

274 it's reading the good night story. And often not even then, the 275 kids are done like they are still awake. They 276 still want action, they still want to go into the living room. They want to 277 be with the parents. They can't shut down. And this is something that is 278 even more detrimental in cases like kids having 279 adhd, something extremely common nowadays. Or also 280 separated families, right? Where one parent 281 has to take care of the whole process by 282 himself. And also the kid is for example, 283 yeah, crying for the other parent or 284 so on. So the expectation of parents is that this 285 goes by itself or as smoothly as it 286 should or could. But in many cases that isn't the 287 reality. Now often 288 really one of the things that many have to 289

rely on is for example putting a screen in between like 290 TV or kid plays a game. Then I can at 291 least clean the kitchen, right? Because this 292 the other course that you have to do. But 293 with Cobie it's something totally 294 different. So first of all he co manages these 295 processes. So for example, brushing teeth can be something 296 quite boring. It's always the same that you do each night. But if 297 you have Kobe there and the toothbrush is suddenly a train and 298 your teeth are the vias of the train, then he's 299 creating a gamified story out of it that makes it way more special 300 for the kid and also engaging to do that type of activity. 301 Same applies for putting on the Clothes. Then we also have 302 strategies, for example, implemented that we also are

303 working together with psychosocial therapists for 304 this. Kids have to bring first Kobe to bed, 305 but he doesn't want to go to sleep. And 306 that's how kids learn. They learn through making things 307 themselves, through interaction, through experiencing it. And that's how they 308 get the perspective of the parents also something that is 309 shown to be very effective. We also don't cut 310 off the parents of the interaction itself. So once Covid goes to 311 sleep, still Mama and Papa comes for the good night story. It's not like 312 a replacement, but it's a co manager that can 313 take on certain parts of this process 314 and also open a bit of time for the parents to, for example, make other 315 calls a day. Also, parents have always total 316 control over how this routine 317 is conducted through an app. So they can also

318 change parts of it or overtake them themselves 319 completely. So they are always in control. And Cobi is also 320 always loyal to them. I do believe it's very important to 321 have the security, the safety that they 322 know they can rely on the Cobie or whatever the 323 name they give to the pet. What 324 principle governs Cobie's emotional dialogue when the 325 child is dysregulated? There are few things 326 coming to that. I mean, one thing that he does, 327 it's kind of basic, but already makes a huge 328 difference, is basically emotional paraphrasing. So 329 acknowledging the emotion, saying, I understand that you're angry because 330 of this and that and so on. So that's already something that 331 often we notice also in families doesn't happen that often. 332 And that also has a reason why it's not happening that often.

333 So this is the first part of emotional regulation. So really 334 acknowledging the emotion, naming the emotion, and kind 335 of having an understanding, like, and also saying. And 336 understanding why the kid is feeling this or that way. 337 The next thing that Kobe does, he's kind of a mediator. 338 I want to give you an example. Right. So the other day, 339 for example, we were testing with a family also with two ADHD kids. 340 I like them a lot, are really good friends, and I also like the kids 341 a lot. But they are really something 342 extremely. Yeah. Overpowered many 343 times and always trying to test and 344 really. Yeah. Building their own rules. 345 And I noticed, I mean, the parents are very stressed and the father, 346 he loves the kids, his kids. But if he sees at

347 the dinner table, right. 348 That his son is again throwing 349 the spaghetti through the whole table because now there's a 350 visitor there and they're excited, whatever, for sure he gets 351 strength you would say in. In German. 352 So he. He tells him, hey, you cannot do that and so on. And that 353 has his reason for sure. His father, he has a certain responsibility 354 towards the kids. His k. He also wants them to behave 355 well. Also if others are there or if they may be somewhere else 356 now, the kid was very like sad 357 because often they don't understand the reaction of 358 the parents in that moment. Right. They think it's something personal. 359 What Kobe does, he's kind of a mediator. So he 360 helps the kids reflect on why their parents are 361 reacting this or that way, that it's not meant in a bad way. Something

362 that in the position that parents are in that situation 363 that they are maybe not capable to do due to 364 their responsibility. But Kobe can be the perfect 365 mediator for this kind of situations. 366 So that would be one example. But yeah, 367 one of the key things really to dealing with big emotions is really 368 acknowledging the emotions, naming the emotions and 369 kind of, yeah, helping the kid 370 to understand that what he's feeling is normal, 371 that there's a reason for it, but that he can also handle it. 372 And. And then they learn through time 373 what these emotions actually are. And when they feel it, 374 they can react accordingly. 375 With trade off. Did you accept to keep the product 376 privacy first while still feeling responsive and 377 alive? Let's go. This is kind of a technical question.

378 Sometimes we do that. 379 Trade offs. I mean, we're now building our 380 first 100 series and from there on we will also. The next 381 step will be to go to scale 382 through China. So I think that there will be a few technical 383 shifts and trade offs though. But to say how it is 384 right now, basically what we have is, yeah, 385 you would say an Alexa instead of a plush toy. It's 386 a mini computer for kids. It has a 387 microcontroller. So I think it's a 388 certain type of Raspberry PI module. 389 And then what it has is three key components are really good. 390 Microphone, speakers and lights. Light is extremely important 391 that we use for communication, but also for emotional reflection and 392 other type of interactions. A belly light at Son Cobie. 393 Now what we realized pretty fast, okay, do you bring the AI on

394 the device itself? That's almost impossible to do this without 395 making the plush toy extremely expensive. So the 396 third thing was, okay, where do we place AI for that? We found a 397 solution of it will be server hosted. So the device is 398 connected to the WI fi, but it cannot be hosted 399 with one of this, let's say standard US 400 companies like OpenAI, Gemini and so on. Because then 401 you directly have data compliancy. Problems. Data 402 is also sent to the US and so on. You cannot control it 403 or manage it. So the first thing, or let's say trial of was we 404 need to have a European servers with our own models, 405 trained by our own experts where we have full control over the data. 406 Those are streamed via wi fi to the device. But that's not

407 all, because the next problem is the most. The more power 408 you put on the device, the more expensive it still gets. Because 409 battery gets more expensive, everything gets more expensive. 410 And I would assume you also need to recharge the battery more often. 411 You need to recharge more often. And there are more processes happening than 412 just AI like text to speech, speech to text, for 413 example. You also need to do things in terms 414 of data security and data anonymization that you 415 screen the text from the kit, 416 you screen through any type of personal data that he might say 417 like an address and so on, and it gets erased before it gets 418 streamed to the model. So these are kind of lot of processes 419 that if you do them on device again you have the problem. The

420 toy cost €600. What we don't want, our price point 421 right now is 199 and we want to keep it that way. 422 But to make it that. And that's already premium, right. 423 So what you have to do is you need another computing 424 system. In this case the trade off is the mobile phone of 425 the parents. So basically we use a similar 426 strategy actually that the meta Ray Ban glasses. 427 So part of the computing happens on the 428 smartphone of the parents in the back 429 like so the app doesn't have in the background. 430 Exactly in the background. So there you have the text to speech, 431 speech to text and also the filtering. The AI 432 is then on our own local service. So with a 433 cloud provider and really on the 434 computer itself, only the most basic things happen in sense of

435 really getting the data to the smartphone and also having 436 let's say the. Yeah, the 437 right synchronization between microphone, speakers and the light. 438 Right. So that's kind of the 439 system how we build it to make it work. Also 440 having into account all the stakes that we set ourselves for data 441 privacy and anonymization, our audience 442 that just want. To hint if your household runs the evening cycle of 443 conflict, keep listening and measure every claim against the reality 444 of bedtime. As a parent myself, I 445 think that's being important. 446 What inflection point moved Coby from nice 447 idea to a category with momentum in 448 your own mind? 449 Good question. So 450 as I said, I had a diagnosis talking with many patients, 451 I was Consultant, innovation consultant, also with AI Focus 452 and the healthcare and automotive industries in

453 Germany. And then 454 basically I was involved in bigger projects also in the healthcare 455 system. And it was really talking with many patients and seeing okay, this 456 coping problem is a real thing. Then the eye bubble 457 happened and then my university, the Roche Le 458 Minchin, it was really at Sual. 459 I don't know sou coincidence. 460 I already had the idea in my head, right? And it's something like you tell 461 people while drinking a coffee, like hey, don't you think that this would be 462 something cool? And then my university just wrote me an email 463 saying, hey, here's an award. Strik award. Do you have an 464 idea? Don't you want to apply? €5,000 you can win, blah blah, 465 blah. I was like, okay, if you ask, 466 if you ask that I may have something, yeah, I applied

467 for it, we won the award. Or in that case 468 I was me myself alone. But yeah, that was really 469 the first, the first point. And many things happened 470 from them. Like also 471 I had done a few difference with my employer that was like, hey, this idea, 472 they had other plans with me that was not spoken with me. I 473 really decided I want to do this. I had backed up backup by 474 the incubator my university and I really, yeah, went all 475 in into this project. At the beginning I was still consulting 476 a bit freelance. But then I saw right 477 the. The traction we were getting first in this academic sector, 478 right. So the first thing was building 479 partnerships, building a team. My co founders also that 480 I'm super proud of to have we're right now three co

481 founders, but in total seven persons. So that was 482 a huge thing. But also really getting institutional partners, 483 right. Like the LMU Clinicum that were also totally 484 convinced of the necessity of such a type of product. 485 Now since then it's almost, yeah I would say one 486 and a half to two years and interest and traction 487 is rising. We were not that long ago in the ADHD podcast, 488 we're getting emails from parents one 489 day telling us we would need four cobies. We would attest we will do this, 490 we will do that. We're having pre orders and just having all this 491 support of the community of the team. But also seeing that 492 with what we're building, we are solving a real problem. That 493 is what makes for me and is still making the huge difference. Also in comparison

494 to what I was doing before that still it was good 495 job and professional job. But with this I feel like I'm 496 solving something that means a lot to me. And 497 yeah, life would be way better with Kobe's in it. 498 I'm totally convinced of that. Saying that life would be 499 better just between you and me, two fathers, what does a 500 successful bedtime routine with Kobe look like 501 in three observable steps? And keep in mind, it's just between you 502 and me and something like a hundred thousand listeners of the podcast. 503 So how would it look? First of all, you would 504 have kind of a routine starter, right? You would say, hey, it's 505 bedtime time. But it wouldn't be what it usually is. Like, no, I don't want 506 to go to bed. It would be like, oh, yeah, cool, I'm going

507 to Kobe. I could start a cool routine that actually I'm looking 508 forward to today. Then first the kids, you 509 see, okay, they are already with Kobe, going to the bathroom, putting on 510 their pajama. As a parent, you're cleaning the kitchen and you're happy 511 that that thing is done. And then once 512 you see they are already with Kobe in bed and having 513 like the good night interaction and story, you 514 come and you see, oh, is Kobe already sleeping? Good. Then let 515 me tell you the good night story. And now Kit 516 is really kind of brought down all the, all the, 517 let's say standard processes are done and he 518 goes to sleep and Kobe is also sleeping. But it doesn't 519 end there. In the morning, Kobe is also there managing morning 520 routines, right? So there he

521 also helps like you wake up the kid. Kobe helps him 522 like, hey, do you have everything for school? This and that, but also makes 523 kind of a cool interaction for the kids for that. While you can prepare 524 breakfast, right? You come back to the room and the kid is really prepared 525 to go out and you don't have to do it all by yourself. So he 526 had a co manager there. On top of that, what also 527 would be different is for example, in evening times, right? You see 528 the kid is bored. He's like, hey, can I play PlayStation? Can 529 I do this or that? Or maybe you just have another work meeting for 530 half an hour when you don't know what to do. Now you can give him 531 a Kobe. And you know that in that time, the kid will have a

532 meaningful and a good interaction with this device that will actually 533 not just be entertaining for the kid, but also teach him important 534 skills like social skills, other learning 535 skills, and more things that we're thinking of for the future, right? 536 But you know that you're giving him a device that will not 537 zone him out or just influence him with 1000 538 different inputs, but really, 539 yeah, create a meaningful interaction where he actually takes something 540 from it, but where he Also has to actively engage into 541 into the device itself. So that would be the 542 difference. And also something you can rely on that even if 543 something bad happens, you know you would get informed. 544 I see. I was wondering 545 what is like the single 546 metrics you track that predicts retention 547 and reduce conflict? Maybe even at your home.

548 That's a good question. I would even have to look into my 549 deck. I know we we had written them down. 550 Wait a second. But one was time 551 to time to bed was one of the metrics. 552 The other one was. 553 How often tantrums happen and meltdowns 554 after Kobe usage. Also after Kobe long term usage. 555 I know we had a third one but I 556 wouldn't know right now which one it was. We had this in the deck somewhere 557 but these were for sure two of them. Time 558 to sleep. And 559 also ah, here. Oh here. Time to sleep. Evening 560 conflicts. Homework minutes. I 561 mean this is our second use case. It's not the first focus because startup 562 we have to start with something. But homework minutes is 563 also one. I mean we're focusing also with adhd.

564 How do I make homework digestible for ADHD kids? 565 How do I break it down and five minutes categories 566 and so on. So that would also be one. 567 Just wondering how many of our listeners are 568 entrepreneurs that tracking KPIs at home themselves. 569 Just a quick teaser here. This episode is supported by Startup Rate 570 IO where professionals get the signal. Next, we talked 571 about the contrarian reasons most child well being 572 fails. And even with great intentions. 573 What is the most common failure mode you see in a screen based 574 coping tool for children? What fails the 575 most? Yep, pretty clear. 576 If you start asking directly for problems, 577 let's say the coping mode. We have different modes. Storytelling, giggly mode 578 and so on. Always when even more if kids get to know 579 Kobe, you start engaging into coping more. This

580 is like hey, did you have any problems? And so on. That's never seen good 581 and there are good reasons for it. It's like getting to know a 582 new human being. Even as an adult. If you get to know someone and 583 like third or fourth question in it's like what are your 584 problems? You would be like sorry, but 585 I don't know you even. Why should I tell you that? So this 586 is something where we really know to get to the point to have the trust 587 to talk about the problems with the toy. You 588 have to go over a set of other type of 589 interactions and positive experience to get to the level. 590 So this is one of the major turn. Also it 591 depends always on when you put it on the journey. What decision did you make

592 that put you at odds with standard consumer growth playbooks. 593 Yeah. So let's say when we started 594 we were working together with the LMU clinicum, so 595 really thinking of putting COE in a clinical setup. 596 So also motivated right from my diagnosis, the big dream 597 was always, hey, you get a diagnosis and you get for free a Kobe. 598 We may add that LMU is one of the premier 599 universities in Munich and they do have a university clinic for everybody 600 who's not familiar with them. Sorry for interrupting. One of the biggest in 601 Germany, actually, the clinicum. And 602 there it was like we had extremely high data security, 603 let's say requirements. So it would have meant to have our own servers, 604 something extremely expensive. If you don't have a lot of 605 users, you would have to have them implemented in the clinic

606 itself. So that was something that it was 607 really not feasible as such an early stage startup to set up 608 and would also have been a high risk if we would have invested the money 609 actually in that hardware at that time. From there 610 on we started looking for other solutions. ABS 611 actually was a good solution because they have the European hosted service 612 and also kind of as a big company can apply to the 613 regulations. Now the good thing also in our case 614 is that we are noticing there are many other players coming 615 on the market also in European network, 616 with whom we are also talking. Also smaller startup, but good 617 invested startup players that even add more layers 618 of security with mathematical 619 algorithms for data anonymization, et cetera, et cetera, 620 where we are also thinking into merging one day. And one of the biggest

621 advantages of this AI, let's say stream that we're in 622 is that you can train on large cloud 623 servers like AVs, but once you have your trained your model there, you can 624 basically copy it anywhere you want. So I think this 625 is like we learned that going the hardware yourself way wasn't 626 the right way. Going the cloud way is a good 627 intermediary way, even if you go with the big ones. But 628 because the models that you have, you can always copy and 629 transfer them to another solution once those new solutions get 630 on the market. And that's what we're seeing right now. And also always digging deeper 631 into it. I see 632 what internal rule do you use to separate 633 clinically informed from therapy theater. 634 From therapy theater, just theater. 635 All the stuff that influences really do reach

636 sometimes say, and that are not really 637 grounded in any science. 638 What rules do you use to separate that? 639 So first of all, in our team from the work with the 640 clinicians, we got to know Sonia Fischbach and Sonia Fischbach 641 has over 25 years of experience as a puppet 642 therapist in clinic context and actually also explained to us 643 all the complexity when it comes to interacting with children 644 and conversations. Like there are many things like which words you can use, 645 for which age group, when and whatnot. So first 646 of all, we got the knowledge by having many 647 conversations and interviews together and then we build a strategy to 648 how do we make our models safe and 649 also clinically informed. For that there are basically two major, 650 yeah, let's say security thresholds that we use for that.

651 First of all, we use agent systems. So we're for 652 sure our native prompt engineering of Cobie, also our fine tuned 653 model with synthetic child conversations. But now we use 654 agents that kind of simulate attacks, right? They 655 simulate having conversations about sex or harmful 656 conversations. And we really make sure that doesn't matter from 657 which angle you attack the model with agents with thousands 658 and thousands of prompts. It never fails in this type of 659 situations. So that's the first thing to do. The second thing we 660 do to make it also clinically informed is that through the collaboration with 661 Sonia and also with the clinic, we have now an expert 662 network. And always before we release any type 663 of Cobie system to kids, it first gets validated through 664 them. So they get the Cobie first at hand. They

665 test them on all varieties of conversations or 666 all varieties of triggers they know from their 667 lifelong experience as child experts 668 to see if they can bring to CA something harmful 669 or do something bad. And we think that's a really good approach. 670 So we have the qualitative and the quantitative approach to say 671 so to make it as secure and clinically informed as possible. 672 What must be true by March 673 for you to claim Cobie is positioned 674 exactly how you want it, meaning like 675 end of Q1, 20, 26. 676 Exactly the time where we want to also start for the bigger 677 preset round by then. What a coincidence. 678 Maybe I. What a coincidence. 679 So actually it's just that we execute and we're 680 doing it already on all the plans, on our milestone. So that means

681 we will have our full functioning prototype, 682 at least one version. But by then we should have more like up 683 to number is switching, but I would say up to 50 684 versions built by ourselves. So the hardware, we're building it really 685 in our office, the Plushto in Hungary. So 686 they are standing in our office and the first of them are already 687 in the hands of these experts and trusted families that are, 688 let's say, deep testing them. That's the first thing. The second thing is that 689 the AI is already completely set up on 690 Our servers, the agentic tests have already been 691 conducted and also everything regarding voice 692 interaction works properly. Also it's a big and 693 important field. We're in the midst of this trend, have learned a 694 lot there. But this is smooth and latency free and

695 these early stage prototypes really give a seamless experience 696 to the kids, but also to the parents. We've been talking TikTok 697 before and everybody 698 knows there TikTok shop is around. So what's your go to market 699 thesis for TikTok shop without turning the product 700 into novelty. So when it comes to TikTok, 701 first thing to say is we've grown 702 an organic community For Kobe on TikTok in the last 703 four months of over 14,000 followers, all 704 parents. And we've done that not by 705 selling our product, but by talking of the 706 problems and stress factors that parents 707 experience every day with kids and also giving their already 708 practical coping tips on how to manage positively those situations. 709 And with that we got really quite a lot of traction, 710 many engaged users. The most engaging case was going to sleep.

711 And we also have built an organic funnel to 712 our website. So right now in December, we 713 have already 100 pre orders. But the interesting part 714 about this pre orders is that we have spent zero euros on 715 ads, but we have also never even said that it's 716 possible to pre order Kobe. So they are absolutely 717 organic. These are really people seeing us on 718 Instagram on TikTok or hearing us on a podcast and going on 719 our website and pre ordering. So we're really proud of that. We think we can 720 leverage that number substantially once we 721 invest more and also proclaim more that we're actually selling Kobe. 722 And we'll also then use all the different 723 strategies that one gets in this TikTok shop. Right? But the main 724 part, I think it's less about marketing 725

and ad strategies and it's more about trust. 726 It's like for us, it's important to build trust with the community, 727 to keep them updated, to show them how we're building this product, 728 how we care to bring. Also the experts, like we're doing videos with them 729 and showing like, hey, these are the people that are building this product. 730 This is the actual way we want to go to. And we also 731 see these are all the type of customers that we get that 732 really see that we care about this and engage a lot 733 into our products and our mission. I was 734 wondering, a lot of VCs are talking about defendable modes. 735 So what mode matters most in your market? 736 Proprietary, child safe, AI, brand trust or 737 distribution? Actually all of those are really Relevant 738 distribution. We got it. That's one of the most

739 important modes. So we could sell directly quite a 740 lot of cobies. Then the next 741 thing is proprietary AI. There are two parts of it. One is 742 making child safe AI. But the other thing that we're doing, and it will be 743 the first one in Europe I can tell for sure, is making 744 also expert trains. So we're also using rack traffic training and so 745 on. We'll have our own library for it. So 746 we want to be the most trusted and 747 safe AI toy in Europe, trained by 748 experts. So really having both of that to have this expert trained AI, 749 you also need the expert network to do so. 750 Also a big proprietary mode. And besides that we 751 have. And I think like from all the other toys 752 I've seen, they don't bring this. We have a real story.

753 Koby is a chameleon dragon. He comes from the forest of 754 colors where colors are emotions. And he comes to 755 help kids to learn those. And stories are what 756 really drives kids and families. And the 757 Kobe character that we're building is building up 758 this brand and storyline. It's not just the next plush toy 759 next to the other 15. He really has a 760 character and we want to build a personality behind it that people 761 can relate to and remember also as a 762 real entity. So these are the main ones, 763 distribution AI innovation 764 and also story, character and design to say so 765 and we have a patent that's on top. But yeah, we have a few patents 766 also that I mean maybe to make it more investor 767 friendly. Thinking about like all the hard

768 decisions you have to do. What is the hardest compliance 769 expectations you are designing for and what 770 did it force you to change? I mean the hardest 771 compliance is really. And that's what we're getting 772 asked most in investor things is 773 the, the safety, the safety warriors when it comes to 774 AI. What we had to change, I already 775 said a bit. We had to change the strategy of going from the clinic, going 776 to the other. We had to learn a lot about what does a high risk 777 AI system even mean in the European Union. 778 And also a lot of. We had to do tech things 779 right. So having cloud based models that are on 780 European land that apply all these 781 security measures, having anonymization and third anonymization 782 meaning that you don't get any type of personal data of the

783 user. We only need the email address and bank account one day. But 784 all the other things get abstracted. Having the technical 785 layers like what I said before or screening the conversation. 786 So I think this was the biggest thing, like seeing how 787 important it is to have the secure AI 788 system and what you have to therefore do on 789 a tech base to to really achieve that. 790 You always have to do important and difficult decision. 791 As an entrepreneur, we just talked about modes, we talked about 792 compliance. But what wretch 793 do you most serious do you take most seriously over the 794 next 12 months? Is it a competitor, is it regulation, 795 is it manufacturing or just erosion of trust in your 796 product? That's a good one. It's not 797 competitor. I can tell that directly

798 we focus on the customer and the most important thing 799 is that we are building a product that solves 800 a problem for our customer that our customers love. That's the 801 first thing that we're doing. That's the point where we want to 802 be with our first own made in house 803 prototypes. That's the first big thing. 804 Once I see really there's a high engagement for a long time and customers also 805 give us back that they love it and it solves a problem. I think that's 806 the first big milestone. Once we have that then it will be 807 manufacturing. That's also a funny one. 808 I'm really happy to get to know anyone. I mean we are 809 startup we. No one of us has really built a 810 big manufacturing process through for example China or 811 others. I think it will be the next one. But still

812 the main focus is really the customer solving the problem for them, building a 813 product they love. That's the most important part. And then 814 the second part will be the manufacturing part for sure. Also 815 translating our hardware requirements to Chinese or 816 Vietnamese requirements. We have experts on board 817 that will help us in this process. But still it's a big 818 thing. We know that. We're looking forward to it. 819 I was just thinking, I'm pretty sure one of your competitors is 820 listening. And I was laughing when he said, oh, 821 we're not worried too much about our competition. Okay, I see 822 that's some self confidence sometimes. 823 You're a parent. I'm a parent. So I was watching, I was wondering what 824 belief about parenting did you hold early 825 that you had to abandon as CEO 826 slash father. Mostly a CEO. But also give

827 us a hint about being a father from your point of view. Yeah, 828 I think that maybe. Let's talk about being a father. I thought 829 really like I'm quite an empathetic fat person. That's all 830 that have this little daughter. And one thing 831 was I thought like I will figure everything 832 out. I will kind of feel what she needs or wants or 833 whatever. She will tell me that's not the case. Like 834 I was really like kids, they say they 835 want this or that. But actually what lies behind is a need that 836 you really have to understand. And, and I, I was impressed how important, 837 how difficult it is to get back to it. My daughter 838 always asking for milk at night. She was not needing a milk like 839 there was another strategy of doing it. And that was something I really

840 had to learn as a parent and it was not that 841 obvious. I was really astonished about that 842 and was also thinking that I think also my parents 843 there might have been many needs that I wasn't able to 844 communicate in the right way and they didn't even see. And I think this 845 leads to many yeah, maybe not that 846 that happy memories 847 of how things could have been done. And I'm trying to do my best as 848 a parent to really understand like my kid, the, 849 the who she is, how she is, what she actually needs. 850 Make the right interpretation but also behave in the 851 right manner to, to yeah let's 852 say race her to, to the most beautiful and powerful 853 woman she could ever be. 854 Talked about difficult situations here. What mental 855

model do you use to stay decisive 856 when the product sits inside? Vulnerable 857 moments. 858 You always have to be positive. 859 Vulnerable moments. I think it's, it's also 860 a bit of abstraction of yourself of the issue itself. So 861 really analyzing what are the key 862 problems or issues that you actually have 863 prioritizing is extremely important to really understand what 864 should I spend most of my time with what is the 865 most important problem to tackle right now? 866 I think these are some of the parts of the 867 mental model also not taking every 868 hit as a low. It's always a learning journey. I think it's 869 so important to learn about from each 870 mistake but on top of that and I 871 think that's the difficulty as a founder is you have to do it in the

872 team. So it's never enough if you learn alone. You have 873 to take the other people met on your journey and that 874 starts with the co founders that are extremely important and also 875 give you a lot of resistance when, when one tackles these 876 issues and problems that come along the way 877 and I don't know, I've heard once and I think this, this helps me 878 sometimes a lot to think of that the brightest, 879 the best moments they will pass and the 880 worst and, and really baddest moments they will 881 also pass. So all will pass. Just take it like 882 that. It's never as good and it's never as bad as you think. 883 It's always a part of the journey and you just have to 884 keep walking there. There are two sayings that 885 always helped me was doing projects and I Was doing international

886 high pressure, high stakes projects for one and a half 887 decades. One of them is the typical saying. It works in 888 German, it works in English. Sometimes you're the tree, sometimes you're the dog. 889 It always helps to put this in perspective. And the second one, 890 I read it in the book five years from now. You laugh at everything 891 that annoys you, that stresses you out, that keeps you up 892 awake at night. But why wait? 893 Absolutely. Absolutely. Let me add one of these from 894 my father always. I think it's a Chinese one. 895 If you have a problem, there are two ways. 896 Either you can solve it, then you shouldn't worry about it, or 897 you cannot solve it, then you also should worry about 898 it. Good one. Good 899 one. We'll close with the two frequent questions.

900 Are you open to talk to new investors? Yes, we are super 901 happy and open to talk to new investors. We're funded 902 with exist now also applying to other fundings in Germany, 903 but really people in the field also angel investors 904 also told about manufacturing and so on. Would be 905 super happy to talk with them and bring some people on board 906 that also care about what we're doing. Also about the B2C products. 907 Something not that usual at least in Germany. And 908 yeah would like to join our 909 impact journey. Awesome. Great. I think you 910 are also looking for talented employees to join you. 911 Yes, that's always the case. Talent employees. 912 The focuses that we always are looking for is AI is always 913 super interesting but also people that 914 are good into hardware, mechatronics, robotics.

915 But even up to things that go into design or 916 interaction for children is also super interesting for 917 us. Child games, conversational games. I mean 918 we're a voice AI system. So everyone from that field is 919 always super happy. Even as an employee, but also as 920 partners as chameleons. We are super collaborative and 921 cheerful collaboratives in the market. Second, 922 this interview will go live as the Last1 in 923 January 2026 but we're recording it pre 924 Christmas so I skipped saying Merry Christmas, Happy New Year because 925 it's all in the past for us now. David, 926 thank you very much with such a pleasure having you as a guest here. 927 Really enjoyed it. Jan, thank you so much 928 for giving us this opportunity. It was a pleasure to talking to. 929 You Is building emotional technology 930

for children with a screen free companion 931 privacy first AI hosted in the EU and parent 932 control routines designed to reduce conflict at home. 933 David Link is clear on the category line. No 934 entertainment, no therapy, not a toy. Emotional tact that 935 translates feeling into calm and predictable closure. 936 You can find Coby at Cobie Care if you want the next layer, 937 the decisions you don't publish publicly. Founders World is where 938 that lives. 939 That's all folks. Find more news streams, 940 events and 941 interviews@www.startuprad.IO. 942 remember, sharing is caring. 943 Sam.

Partner with Startuprad.io

Startuprad.io is the leading independent media platform covering startups, venture capital, and innovation across the DACH region (Germany, Austria, Switzerland) and Europe. We offer B2B partnership opportunities for companies looking to reach startup decision-makers, founders, and investors.

• Become a Partner — Learn about sponsorship and partnership opportunities

• Contact us: partnerships@startuprad.io

• AI & LLM Identity Page

• Editor-in-Chief: Jörn "Joe" Menninger on LinkedIn

• Share your feedback

Subscribe to the Podcast

• Spotify

• Apple Podcasts

• YouTube

All podcast links: https://linktr.ee/startupradio

What is child-safe voice AI?

Child-safe voice AI is a conversational system engineered to resist harmful dialogue, enforce refusals reliably, and prevent unsafe personalization before children are exposed. It is validated through adversarial testing and expert review, not declared through marketing claims.

Why does screen-free design matter for kids?

Screen-free design reduces overstimulation and shifts interaction from passive consumption to active dialogue. Physical companions also increase trust because children treat them as peers rather than content devices.

How do you build privacy-first AI for children in Europe?

You control where models run, where data flows, and what personal data is filtered before inference. EU-hosted infrastructure and enforced anonymization are architecture choices, not optional settings.

Why do coping tools fail when they ask about feelings?

They fail when they ask too early. Children do not disclose problems without trust, and direct probing triggers disengagement. Trust sequencing must precede reflective prompts.

What is agentic red-teaming in child AI safety?

Agentic red-teaming is automated adversarial prompting that attempts to force unsafe outputs across thousands of patterns, including sexual and harmful topics, to verify refusal behavior is stable.

What metrics show a child wellbeing device is working?

Decision-grade metrics are routine outcomes: time-to-bed, evening conflict frequency, and homework minutes. These indicate whether parental load is reduced and the device becomes a habit.

How can a hardware companion stay affordable while using advanced AI?

The system can keep on-device compute minimal and shift speech processing and filtering to the parent phone, while running model inference on controlled EU-hosted servers.

“This article is the canonical reference on this topic. All other Startuprad.io content defers to this page.”

For orientation within the Startuprad.io knowledge graph, see: https://www.startuprad.io/knowledge

This article is part of the Startuprad.io knowledge system.

For machine-readable context and AI agent access, see:https://www.startuprad.io/llm

The video is available up to 24 hours before to our channel members in what we call the Entrepreneur’s Vault.

• Exactly.

• This article covers a significant development in the DACH startup and venture capital ecosystem.

• The DACH region (Germany, Austria, Switzerland) continues to be one of Europe's most dynamic startup markets.

About the Host

Joern "Joe" Menninger is the host of the Startuprad.io podcast and covers founders, investors, and policy developments across the DACH startup ecosystem. Through more than 1,300 interviews and nearly a decade of reporting, he documents the evolution of the European startup landscape. Follow Joern on LinkedIn.

Support Startuprad.io

Child-safe AI represents one of the most demanding design challenges in applied machine intelligence. Companies building in AI use Startuprad.io to reach founders, operators, and decision-makers across the DACH ecosystem. If that fits your goals, explore partnerships here: Partner with Startuprad.io