Building a real-time AI tutor for 5-year-olds
ello.com31 points by catalinvoss 4 hours ago
31 points by catalinvoss 4 hours ago
Hey HN! We've spent the good part of this past year building an AI tutor that teaches kids ages 4-9 reading, math, ESL and more. Getting an AI tutor to effectively teach a child turns out to be a really hard technical challenge, this took getting the underlying architecture right.
Our tutor steers the UX in real-time and makes complex decisions on the fly. Doing both at conversation speed required us to replace the standard tool-use loop. We built our own tutor harness that utilizes a streaming interpreter that executes actions, while an asynchronous planner model reasons ahead of the conversation and makes calls that drive the child's learning. On top of it all, we developed a safety system that checks every turn without it causing an interruption to the activity and conversation flow.
Effective teaching isn't just about answering a child's question quickly, rather making the right move at the right moment. AI is also going to be an integral part shaping how this generation of kids learn to read and think, tackling this responsibly means getting the design right.
Happy to answer questions and curious what you all think, critical feedback included, we've been working on this problem for a long time and love to hear from the HN community.
This is so cool! Ignore the haters, AI accelerated education is so obviously a gigantic win for everyone. (And massively levels the playing field.) I wish I had had this when I was a 5-year-old. Few of my teachers really understood the things I wanted to learn, my peers weren't interested in the nerdy things I was, and my parents certainly didn't have the wealth to provide me with private tutoring. There are a lot of negative comments here, but they are shallow... I'm sure those commenters wouldn't want to live without the access to the Internet, and even a brilliant five-year-old can't use the Internet effectively yet. A smart and curious 5-year-old has endless questions and a properly harnessed LLM has endless patience to provide answers at a level the kind can understand (which usually not even it's parents do). In fact, this could be one of the most beneficial uses of AI for society yet... private tutors of the level that the mega-rich always had, now for all kids everywhere! This gives me real hope for the future generations of humanity. I was that same 5 year old. I do think that if we want AI to force-multiply humanity, we need to start leveraging it for education. I think it's one of the biggest levers we have to be honest The reality is, most 5 year olds don’t get access to the resources most of us have had while growing up. People are saying, “kids should have human tutors.” Guys, most people in the world don’t have any tutors! What Ello has built and other forms of AI-based tutoring is going to raise the average level of education and literacy in the world. Especially in developing countries. Let individual parents decide what’s best for their kids. What are your thoughts about children in a Sudbury School model? These are democratic schools where children can do what they like in the day. Mostly they choose to play with other kids, games of imagination, though also doing screen time. One of the basic principles is that children figure out what they want to do and the learning comes along with it; the model views adults wanting children to learn something specific as generally counterproductive though having resources available is okay if it is not coupled with any expectations. Are your devices likely something that they would have fun with and choose to engage with or is it likely to be ignored unless adults use some kind of persuasion to have them use it? Is it cool with a child using it for a bit and then not using it for a few months and then wandering back to it? How far up into math does it go compared to what an a randomly sampled adult could actually do mathwise? Also for reading, are you using phonics or whole word sighting? For math, to what extent is it screen manipulatives versus manipulations of digits? Also, do you have provision for an older child to start learning this stuff so the basics need not be at a 4 year old presentation level, but the concepts still need to be covered? In Sudbury schools, the typical age of self-taught reading is 7-9 though it can range from 4 to 12. Useful arithmetic usually seems to happen much earlier than reading though reading tends to get completed by the children on their own while arithmetic does not advance further than the needs of money exchange without special effort. In the long run, Sudbury students have no problem with college level material, including mathematics, but it could be nice to have something that eases the white knuckling if it does not undermine the child's self-directedness. Hi! I'm Elizabeth (one of the co-founders of Ello). This is an interesting question. I actually think there is more overlap than people might assume, but it's a bit more because adaptability to various approaches to learning is part of the point. While I'm not deeply familiar with the Sudbury School model, I think there are various approaches to teaching kids that are successful because different approaches serve different types of kids and learners. I can see why this approach would be successful for a certain profile of student for whom it's the right fit. We start from the belief that children are naturally curious. Our job is to build something engaging enough that a child wants to interact with it because it is interesting and rewarding. If a child in a Sudbury environment never chose to use it, I would see that as useful feedback for us, not a problem with the child. There are opportunities for kids to explore and incorporate their interests within our app. I also think it is completely fine if a child uses it for a while, disappears for months, and comes back. Learning is rarely linear, and technology should be able to pick up wherever the child is. On reading, we’re firmly grounded in the science of reading, so we teach through explicit phonics rather than whole-word memorization because that is best practice. On math, we’re much more interested in helping children build intuition and conceptual understanding than simply getting answers. AI gives us the flexibility to use conversation, visual models, stories, or symbolic math, depending on what helps a particular child understand. One thing AI is uniquely good at is meeting learners where they are. A 10 year old who is learning to read should not have to work through material that feels like it was made for preschoolers. The underlying concepts can stay the same while the language, topics, and presentation become age appropriate. I don’t think there is one educational model that works for every child. What excites me is that AI makes it much more feasible to adapt to individual learners instead of expecting every learner to adapt to the same experience. Super curious to hear from the parents here: Honestly, at this point isn't not exposing our kids to AI just setting them up to fail in the future? Like not letting them learn to use the internet? I have friends who are actually teaching their kids how to use AI because they don't want them to fall behind What does teach how to use AI mean for grade schoolers? It’s a search engine for those purposes, the lesson is a blend of the Google and Wikipedia lessons. No like in general not just for school - one of them taught their kid how to create their own bedtime stories, like with illustrations. Another one taught their kid how to make their own games with AI. Though yes I also literally have friends who have put their kids in alpha school which is SUPER ai. Like since i'm sure most of us work in software, how is this different from letting your kid learn how to code early. Wouldn't restricting access just make them...more illiterate with AI My oldest turns 6 in just over a week and my initial reaction to this heading - as well as the product itself and the picture of the kid using it - was heartbreak and sadness. Not anger, just sadness. Like when you read a story about a kid that's a victim of a crime. Stepping back, I can look at it somewhat objectively and see that there are both kids that need something like this and that it's probably a better solution to the "dumb" homework apps that the kids use for 20 mins a week at this age, but I don't think "Ello deprives 5 year olds of human contact" is the message you should be putting out into the world. I deeply wish I had AI-accelerated education growing up rather than sitting bored out of my mind while a teacher lectures at the bottom quartile I'm a mom who actually has kids and this thread is insane. 'Just get a tutor'...okay?? Are you paying for it? Because that's not an option for a lot of families. I get that it's more ideal, but the alternative is...nothing? Do you not agree that all kids deserves a chance to read? Are we not seeing the lack of reading proficiency in the majority of American adults nowadays?? Or yall too stuck in your tech bubbles?? There are high school students graduating who cannot do math. This is tech that is actually being used for GOOD here. Full disclosure: I worked on a small project with Ello / Catalin a few years ago. At the time of writing, the sentiment in this post is that this is a terrible idea, and that kids need human tutors. This is 100% true. But also, you might want to know some facts about the state of US literacy (Ello is a math and reading tutor): 1. We're in a children's literacy crisis. As of 2025, 40% of fourth graders are reading below basic levels [1]. 2. There's a massive teacher shortage. 2025 US state data shows ~400k teacher positions either unfilled or underqualified [2] – over 10% of the workforce. 3. Bloom's 2-sigma shows that 1-1 tutoring delivers outcomes at the 90th percentile of classroom teaching. Early research is finding that AI can deliver some of this benefit [3]. 4. This can't always be solved by parents: 54% of US adults have a literacy below a 6th-grade level, and 20% are below 5th-grade level [4]. At Ello, I heard stories of children figuring out they were behind at school, and when given the app, they holed themselves up in their room and used it to get themselves caught up. And then they could read! Can you imagine falling behind at this critical juncture, and being stuck illiterate while your friends grow past you? We're setting kids up for lives of shame and deprivation. My take: this really is a life-changing technology. And we need this problem solved. Democracy doesn't function without an educated population. [1] https://www.nagb.gov/news-and-events/news-releases/2025/nati... [2] https://learningpolicyinstitute.org/product/overview-teacher... [3] https://www.sciencedirect.com/science/article/pii/S2666920X2... [4] https://www.thenationalliteracyinstitute.com/2024-2025-liter... Pretty interesting. Hopefully this ends up being an affordable solution for without the means for hiring a human tutor. Dear God no. Keep kids away from AI, and keep AI away from kids. Kids need more human contact, not less. The more I think about it, the more I want to ban your entire business model Agree that kids benefit from more human interaction, not less. That's our goal. The reality though is that the traditional school setting doesn't provide for that: a teacher in front of a 30 kid classroom can't cater to every child and it's not a particularly interactive experience. The current system just isn't working: 60% of US fourth-graders are behind in reading, 40% lack basic literacy. Those kids are going to move on to the next phase of school without the skills to thrive. There are 270 mil kids out of school globally. So what are you going to do? Give every child a 1-1 human tutor? For sure, if you can, that’s amazing. But you can't pull that off. You don't have enough teachers. Technology gives us the opportunity to catch kids up. By doing that, you can decouple teaching hard skills and free up teachers to focus on the things that are truly human and unlock a lot more people who may not have the skills to teach the full curriculum themselves to act as learning facilitators. That leads to more human interaction. Business model is interesting BTW. We’re a public benefit company and are planning to make our products completely free in emerging markets and have a free tier in the US. Impact and business should be aligned here, but pulling in opposite corners. Really awesome work! I've been trying to do some of this real time back and forth voice coaching myself and it's no easy feat. Congrats on the progress. Yeah – this is hard stuff. If we're successful in getting the this right, our users won't be feeling any of the complexity. Please don't. Don't deprive children of the interaction with other human beings. 5 year olds don't need tutors. They need play, touch, sense, feel, run, breath, sky, earth. We agree with this sentiment. Our goal is to supplement children with effective learning, NOT replace humans. This article can provide a little more context on how we're thinking about this: https://www.ello.com/blog/ai-should-make-clear-what-reality-... I’m curious why it has to be an either or? Spending 30 minutes with a tutor doesn’t deprive children of interaction with humans. If we can support a child’s learning (perhaps even more efficiently) doesn’t it give them more time to do that? Replacing a human tutor with an AI that will wildly hallucinate is wrong, and will directly contribute to the dumbing down of society that people here often bemoan. This is the equivalent of "parenting" by putting a kid in front of YouTube Kids for half the day "That will wildly hallucinate" feels wildly hyperbolic considering how fast models are improving in their abilities, especially with proper safeguards built into the model harness. Some 5 year olds could use a human tutor. Giving them AI instead is no different than plopping them in front of "Youtube Kids" instead of being a parent. "Some 5 year olds could use a gourmet meal. Giving them a mass-produced TV dinner is no different than sedating them with opiates* and pouring garbage on them instead of being a parent." * the more things change the more they stay the same: https://blog.sciencemuseum.org.uk/the-addictive-history-of-m... "Don't be snarky." "Please respond to the strongest plausible interpretation of what someone says, not a weaker one that's easier to criticize. Assume good faith." Also, please don't use quotation marks to make it look like you're quoting someone when you aren't. That's also an internet snark trope and we're trying to avoid that kind of thing here. p.s. That's an interesting, and heartbreaking, historical link. Is your hallucination rate 0.00000000? If not, then it doesn't deserve to be used. Modern AI needs to go away. You're not helping by making something that will be grossly misused once it's out of your hands. I get where you're coming from. There’s a lot of potential to misuse AI with kids. What we do believe is that children will be living in a world where this technology will exist, and how it gets used becomes the important question. We also have to prepare them for that world and how to thrive in it. I would never give my son raw ChatGPT the same way you wouldn’t give a 5 year old access to the raw internet. But that doesn’t mean that the internet can’t be used for learning. We don’t have all the answers and we can’t respond for all of AI, but we’re a team of parents, teachers, and child psychologists who deeply care about getting this right and unlocking the opportunities for kids. The article goes into the technical depth of how we make it pedagogically aligned, safe, non-slop. What is being taught to 5 year olds? And why would an AI tutor be better than an pre-k learning app Most students are pretty homogeneous in learning at that stage We started with reading, providing patient coaching as kids learn to read out loud. We are now adding math and in some countries, English as a Second Language. Students actually aren't as homogenous as you might think. And it's one of the big challenges teachers have with a classroom of 25+. They're forced to teach to the middle, which isn't great for kids that are slightly behind or ahead. An AI tutor has the advantage to adapt and teach to each child's unique learning path, make sure core concepts are covered on an individual basis before moving on. Have you read the Sal Khan thoughts on AI and education? About 1-2 years ago he had similar thoughts to solve that exact problem you mentioned. https://www.chalkbeat.org/2026/04/09/sal-khan-reflects-on-ai... Sal khan being the founder of Khan academy the most popular online education course Yes! Deeply admire what Sal Khan set out to do. One of the original pioneers of how technology could transform education. What we learnt from it: a chatbot is not enough to teach a child though. We need more to fully engage them and have the tools and context to truly teach them. We describe this in the blog post, curious what you think. Being frank I think its not enough to have a good looking app and have some llm calls Think about your competition for your market. When I want my child to really excel in learning, I would force them into kumon - so they can skip a grade. If your a student who wants to learn you have khan academy. And im just not seeing anything that screams-this is better than khan academy and kumon All i see is an education app with good design Sorry if it sounds harsh P.s. if youre on the mission of educating people from developing countries-different story and different problems. Ignore what i put here then Please define "we". > Students actually aren't as homogenous as you might think. And it's one of the big challenges teachers have with a classroom of 25+ True. It's well known that some % of students do well with individual tutoring. Move faster, understand things better, etc. And another part of students don't do well with that. They need other things. Maybe help from their peers in smaller groups (like 3..8 students), some after-school extra, a fix for problems back home, whatever. But 5y olds? They need contact with peers, play, attention from humans, run around, build stuff from Lego blocks, touch grass, etc. Learning to read, "3x4=12" math etc isn't hard enough to warrant putting 5y old kids on AI tutors. we = a team of teachers, AI experts, child psychologists, learning designers, and parents; building in the US and Kenya It's bad enough that schools give 5-year-olds tablets to do their maths work on. Let's not expose them to AI brainrot now too. Ai is brainrot because that’s what many people choose to produce with it, and it’s easy to produce brainrot at scale with AI. But it’s hardly the only thing you can produce with it.
Crap content is definitely over represented. It’s an error, though, to think that is all AI is capable of. If quality is the goal, and you are willing to invest the resources to achieve it, you can easily create very high quality work. But it’s not terribly easy. And it’s not terribly fast. It is relatively cheap, maybe 1/4 to 1/10 the cost of doing it with qualified humans. But it’s not trivial and it’s not magic. It’s a force multiplier, but the quality of the idea and the performance of the model used are very important, and good models cost money to use… about $50-100 an hour if you are really leveraging it. But you can do ten hours of work in an hour or two. This sounds like a terrible idea. 5 years olds need human interaction with a human tutor or with other children. Human tutor is definitely the best thing we could do, but it’s not attainable to give every child a human tutor and clearly the current system isn’t working. If you manage to build a good AI tutor, you can unlock more human interaction outside of self-directed learning time oh boy, who validates that the blackbox of bullshit is spewing valid information and not the typical nonsense ...?
teichman - 7 minutes ago
jbotz - 2 hours ago
catalinvoss - 2 hours ago
jordanmeyer - an hour ago
jostylr - an hour ago
eba7keb - 14 minutes ago
ooopsnevermind - an hour ago
Yossarrian22 - an hour ago
ooopsnevermind - 15 minutes ago
AussieWog93 - 9 minutes ago
teichman - 4 minutes ago
frantang3 - 29 minutes ago
dcx - an hour ago
coup1799 - 32 minutes ago
LocalH - 3 hours ago
catalinvoss - 2 hours ago
catalinvoss - 2 hours ago
ilyausorov - 4 hours ago
catalinvoss - an hour ago
28304283409234 - 3 hours ago
catalinvoss - 3 hours ago
eba7keb - 3 hours ago
LocalH - 3 hours ago
willmarch - 35 minutes ago
LocalH - 3 hours ago
tekne - 3 hours ago
dang - 3 minutes ago
LocalH - 3 hours ago
catalinvoss - 2 hours ago
JimsonYang - 3 hours ago
catalinvoss - 3 hours ago
JimsonYang - 3 hours ago
catalinvoss - 2 hours ago
JimsonYang - 20 minutes ago
RetroTechie - 2 hours ago
catalinvoss - 2 hours ago
ErroneousBosh - 3 hours ago
K0balt - 2 hours ago
AlexeyBrin - 3 hours ago
catalinvoss - 2 hours ago
nickphx - an hour ago