▲ 102 points

Launch HN: Bloomy (YC S26) – AI-powered mastery learning for K-12

by alexsouthmayd·27d ago·107 comments·view on hn ↗

Hi HN, I’m Alex Southmayd, the founder of Bloomy (https://bloomylearning.com) – an AI-powered mastery-learning platform for K-12 students. Bloomy provides students with an AI tutor alongside adaptive curriculum (right now Math, English Language Arts, and Writing).

How it works: we diagnose students’ skill gaps, place them on personalized learning paths, and give them standards-aligned lessons and a Socratic AI tutor that scaffolds their learning without just giving away the answer.

The goal is to solve the Bloom 2-sigma problem (https://en.wikipedia.org/wiki/Bloom%27s_2_sigma_problem) with AI.

Short launch video: https://tinyurl.com/bloomylearning

Longer product demo: https://youtu.be/XHvoKt6qMeo

Families access for Bloomy: https://bloomylearning.com/families

I started as a teacher. I taught 7th-grade English and writing with Teach For America, and every day I struggled to deliver differentiated instruction to 30 students with 30 different sets of needs. Some students needed remediation, some needed acceleration, and many needed a tutor sitting next to them helping them reason through the next step. Benjamin Bloom’s two-sigma result—that one-on-one tutoring can produce much better outcomes than conventional classroom instruction—always felt intuitively true to me. The hard part was making that kind of attention affordable and available to every child.

Then AI changed the cost curve. When I saw schools such as Alpha organize academics around mastery rather than seat time, the model clicked. If you’ve heard of Alpha School, that is directionally the kind of learning model that inspired us. But I kept thinking about the families and schools that already exist: homeschool families, microschools, hybrid schools, and regular classrooms where most children are today.

Most students and teachers see learning gaps at the wrong resolution. They get a grade, percentile, benchmark score, or broad standard—not “this is the next skill this student should learn.” Existing personalized-learning products often feel like digital worksheets: they provide plenty of practice, but not much diagnosis or teaching. Very few have AI tutors providing the core instruction. Bloomy starts with a diagnostic—we integrate with third-party assessments and provide our own—and creates a learning path for each student. Students work one skill at a time, receive a short lesson, practice at an adaptive difficulty, and only move forward after demonstrating at least 90% mastery. The learning path updates as the student works, based on their performance and our knowledge graph of skill prerequisites (built in collaboration with Learning Commons / Chan Zuckerberg Initiative).

Each skill has three stages. Base Camp teaches the concept with worked examples. Climb provides guided practice and Socratic support. Summit is an independent ten-question mastery assessment with no hints or AI assistance. Students need to achieve 90% on the Summit to advance. If they struggle too much, they’ll be routed to a different skill better suited for their level.

BloomyBot is not a blank chat window but rather a live, interactive, and observant digital tutor. During practice, it receives the active passage or problem, the question, the student’s attempt, an authored explanation, and relevant misconception context. It follows a scaffolded tutoring ladder: first asking what the student tried, then pointing toward the concept, suggesting a strategy, working through one step together, and only providing heavier scaffolding after the student has struggled, adapting to and learning from the student along the way. Students can interrupt it, and we’ve begun to roll out multilingual support for Spanish, French, and a few other more niche languages that customers have asked for.

We currently use a variety of Anthropic and OpenAI models for BloomyBot. The tutor is restricted to the current lesson, redirects unrelated questions, limits conversation length, and is unavailable during mastery assessments. The language model does not choose the curriculum or decide whether a student has mastered a skill.

That separation is important. A conventionally “helpful” AI response can be a bad tutoring response: if it gives away the answer, the student completes the task but may not learn anything. Our goal is not to build a homework-answering chatbot. It is to put AI inside a structured loop of diagnosis, instruction, practice, feedback, and independent mastery.

LLMs can still be wrong, and we do not claim our constraints eliminate that. We reduce the surface area by grounding BloomyBot in authored lesson content, keeping it on topic, logging conversations, and removing it from assessments. Teachers and parents can review tutoring activity, students can report problems, and safety signals trigger human alerts and a backup audit. We also do not see Bloomy as a replacement for teachers, parents, or human tutors. A good human tutor is better. The narrower question we are testing is whether, during a bounded learning session a student would already be doing, a context-aware tutor can provide better help than static “correct/incorrect” feedback. Longer term, the question becomes more whether a student would perform better with one-on-one AI tutoring (at least in certain aspects of the curriculum) than with many-to-one instruction in a medium- or large-sized classroom.

Bloomy is now being used across several settings: traditional districts, charter schools, hybrid schools, microschools, homeschools, and families looking for additional academic support. In an early pilot at a charter school in Massachusetts serving ~150 students in grades 6 through 8, students averaged roughly 1.8 times the expected winter-to-spring NWEA MAP growth. This was an observational pilot, not a randomized study, so we treat it as an encouraging signal rather than proof that Bloomy caused the difference.

Parents and teachers can see what a student has mastered, what is in progress, and where support may be needed. We have found that adults generally do not want another generic score; they want to know which small number of skills deserve attention this week.

Bloomy makes money through family subscriptions and school licensing. ELA costs $39/month or $279/year per learner, and Writing Studio costs $19/month or $139/year. Math is scheduled to launch July 31 at the same price as ELA. Schools and microschools pay per student, with pricing varying by subject coverage, enrollment, rostering, and implementation needs.

Because children use Bloomy, we collect learning responses, progress data, and tutoring conversations. We do not sell personal information, use child data for behavioral advertising, or permit model providers to train general-purpose models on identifiable child data sent by Bloomy. We have Zero Data Retention agreements with both Anthropic and OpenAI. Parents and schools can request access, export, correction, or deletion under the applicable account or school agreement.

More background on me: after Teach For America, I taught and designed GMAT and GRE curriculum for Manhattan Prep / Kaplan, led the driver acquisition team for Lyft’s New England markets, completed an MBA at Stanford, and led AI transformation projects at McKinsey (so when models finally became good enough this past January to achieve the kinds of things I am pursuing with Bloomy, I was in the right place at the right time to begin building). Bloomy brings together the different parts of my career that I care most about: educational outcomes, learning design, building products, and getting useful technology into people’s hands.

I’d especially value feedback from parents, teachers, engineers working on child-facing AI, and people who have built tutoring, assessment, or adaptive-learning systems. Does the separation between guided AI help and independent mastery make sense? Where do you see the greatest potential with AI in education? Where are our safeguards insufficient? What evidence or product behavior would you need to trust something like this with a student?

Certainly there are many dangers and pitfalls we must beware of, too, but I believe we can really move the needle in K-12 (for the first time in a long time) if we use AI responsibly and intelligently.

107 comments

I am here to tell you HELL YES and TAKE MY MONEY. I am such a fan of the idea of using AI to help give personalized and structured AI lessons in the hands of students and let them cook!

- It looks like you are gating family access to K-3 for now and I think that's right. I wouldn't really be comfortable giving my first-grader a live chatbot. Maybe I would think about whether there are other non-persona modalities that could still be self-directed (i.e. I am uncomfortable with a chatbot interface on this for a six year old but gamified flash cards with options could be different).

- I think the other issue is with motivation. I have various duct-tape versions of these types of agents and the thing about it is if you're doing the learning right it can be HARD. So I would think about using motivational interviewing or other techniques to help keep the user coming back and motivated.

- I would really think about the assessments here too. Many people are worried about LLMs ruining student evaluations, but if you could bake in reliable, flexible exams that gauge user progress (even for something like a "Did you read this" quiz) I would bet teachers would like it. There is likely so much you could do on student progress observability and e.g. structuring team-based projects or having targeted student working groups to hash out hard concepts in a targeted way, etc.

This is such an interesting market and use case too because the educational system might be very structurally set up to the current pedagogical staffing model (think about the incentives for teacher's unions and administrators). If you think it will be hard to change that system as quickly as you want I would also try to have an offering direct to families / home-schoolers. I think there is also a cottage industry of tutors that might benefit. Maybe partnering with the textbook publishers? I'm sure there are some "Teach your kids better" influencers that would get you into some feeds?

Thanks for the support!

On motivation: we have something called Bloomy Bucks. Kids earn these for getting questions right, mastering skills, and being consistent. They can then redeem them for privileges that the teacher/school/parent sets (e.g., homework pass, game time, whatever). This is part of the effortful dopamine that I, personally, think is the right approach to motivation. A lot of people will disagree with me on this, but I think that some amount of extrinsic motivation (something we've already been doing for quite a long time with letter grades) can cultivate intrinsic motivation. One specific application of this is that students can earn up to 1 Bloomy Buck for a meaningful interaction with BloomyBot on a given question.

On assessments: imagine if we could get rid of assessments and just have live diagnosis happen all the time? With enough data from practice and mastery, that should be possible.

We have an offering for homeschoolers, indeed. Bloomy is reimbursable in ~15 states right now through ESA-type scholarships. Great ideas!

Other thoughts:

- I think the other mistake I see here is trying to over-engineer a deterministic learner path instead of giving the AI more free reign on best next interaction and a set of goals it needs to accomplish through the session; it can feel more responsive and free-form that way from the learner's POV. - If you had voice here you could also make the screen optional. In my experience typing out long answers to questions can take a while too - so a voice mode might be helpful for learners. It would also be cool if people could take a 'photo of their work' for e.g. math equations done by hand. - To the earlier point on family end-market, an interesting idea is modeling bloomy - have some grown-up oriented courses so you can learn with / side-by-side with your child? Just an idea.

So, professional educator here, with fifteen years of high school teaching experience. If the sample passages and questions on the front page accurately reflect what's inside this, I would not let my kids go anywhere near it.

You've chosen to use the most generic AI-generated prose, when you could have licensed or hired writers to craft thoughtful, rich passages. I doubt you've consulted any serious standardized test item writers in the construction of what is essentially a standardized test item-based curriculum. This is the pedagogical equivalent of a Lunchable--mass produced, wrapped in plastic, and basically nutrition-free.

100%

This looks very poorly thought out by the site alone. The quotes are all concerning as well because all three of them have emdashes. Maybe they're paraphrased, but then those aren't the actual quotes from those educators.

Parents and teachers who are willing to put middle schoolers in front of screens more are doing our future a disservice. There are real studies (none that I can find on Bloomy) showcasing how this negatively impacts students [0]. I doubt you'll find a correlation that more screen / compute interaction at younger ages improves long term cognition as it's not how kids develop.

Beyond this every parent should run after a glance at the privacy policy and the sheer number of different backend services required to run this. This is a privacy nightmare, long-term, for kids who have no choice. School districts, and a lot of parents, don't understand any of this. I don't even see how this abides by laws in some states simply because there are no protections of model training by some of the examples provided in the privacy policy. This does not abide by our state laws requiring a provider to be able to remove student data permanently within 90 days. In fact no LLM that doesn't have a "no train" clause abides by this simple requirement and it's unfortunate that districts will get sued over using tools like this in the near future.

My advice to this is: if you want to build a tutor then target parents. When you try to shove this down the throat of every student in a district you're making a lot of assumptions and impacting many negatively long term. This may be great for disadvantaged areas, but this is not how kids need to learn. They need to learn, explore and make mistakes on their own and not be preconditioned to the easy answer.

The iPad generation of parenting is about to get a whole lot darker as more and more of these types of companies are taking shots at easy targets, like school districts, with little to no studies behind impacts.

Finally, very expensive! I ran a simple 5 student cost calculation and it's over $2000 a year for those students. And that cost does not include math which is currently not available which would be another $1400.

[0] https://jamanetwork.com/journals/jamapediatrics/fullarticle/...

"professional educator here, with fifteen years of high school teaching experience"

You do ACT / SAT prep and tutoring, correct? Just wanted to be clear, because the description here sure makes it sound like you're a high school teacher.

Thanks for engaging! Our passages are a mix of public domain passages and in-house authored material which are specced to a long list of requirements. This allows us to create passages spanning a rich variety of characteristics, including genres, contexts, settings, cultures, socioeconomic backgrounds, household family set-ups (e.g., single-parent households, grandparent-led households), etc. We also rigorously vet our content using AI evaluators from Learning Commons, which were created using a golden set of inputs from a panel of experienced educators.

I also have several years' experience writing standardized test preparatory materials. Ultimately the reading comprehension modules serve the end goal of teaching students discrete skills, which roll up into standards. Those standards are (as you point out) the main focus of standardized, summative tests.

We encourage students to read real books! We have an AI writing tool that integrates with schools' novel studies curricula to generate a wide variety of prompts so that students can engage in long form writing about the books they are reading.

Article Title: Ed tech is profitable. It is also mostly useless

https://www.economist.com/united-states/2026/01/22/ed-tech-i...

Congrats on the launch! I'm curious if you're familiar with ED Hirsch's work and if so how you see a "core knowledge" curriculum fitting into AI-enabled tutoring. Your landing page suggests you might be falling into traditional anti-patterns for strengthening reading skills like "finding the main idea" and "close reading" of the text (your ai tutor animation instructs a student to look closely for the answer). These patterns are commonly used in American education and have largely been discredited. They persist due to outdated learning theory taught in American educational schools and are likely responsible for our falling reading scores. I'm bringing this up because my hope is that any ed tech tool regardless of whether it uses AI would start with the right foundational theory.

First of all congrats on the launch.

I worked at an edtech startup with adaptive learning platforms in previous decades. One thing that is always difficult to surmount was that at the under K12 level were:

1. B2b it was hard to land large contracts(like getting an entire public school district to jump from large publishers like HMH or Wiley) because it turned out that the teachers themselves did not care about adaptivity when they had a thousand other things to worry about. But teachers are not the ones that make the deals. It's the school district board and The reps at large publishing just said we are also adaptive and no one would care about how you differentiated. Teachers and students really aren't the lot you have to convince in a b2b deal do you :)?

2. If you want the b2c route then the customer acquisition cost was way too high because again like to get teachers and students to actually adopt this to help them was really a hard sell, especially when the public school system has teachers that are so overburdened with other stuff.

I see that you mentioned charter schools for your pilot, which are not exactly public schools but I dunno of they come with the same set of challenges, but I wonder are these above two challenges are Also something you have had to face while scaling your product?

OP this is it

Almost all the founders who have tried to go into education have come out with similar thoughts.

Schools are basically legacy companies with low nps products. It might do you better to vertically integrate as hard as that is

Hey first of all, Congratulations on the launch. I'm a teacher at a Dutch applied university where we use mastery learning to teach students software development, the code for our learning system is open source so feel free to take a look [1].

I personally always followed traditional schooling but programs like flip the classroom or https://www.khanacademy.org/ really helped me and I now also see the same value in my own teaching and your work so thanks for working on this.

At my program we have experimented a lot with teacher bots, AI grading, using AI to generate assignments and other forms of AI but in the end we always came back to 1on1 feedback sessions and ditched the AI to be just an optional tool. In a teacher feedback session a teacher would open the work of the student and grade the work based on how much of the objective was reached. In our experience students will find the tools to not do there homework and that is more than fine as long as they understand what they are doing. We eventually settled on 1 or 2 day exams without AI. Screen recording as proof. Where you get graded on the same objectives and can use any trick you have learned. Assignments typically take 1 day to compleet. Assignments need to be redone if a student scores less than 80%. On the exam often 55% is a passing grade.

So some of the things that pointed out to me where: 90% correct sounds really high. A lot of AI access sounds like a lot of trouble shooting to find the right amount and per student this amount may be different. We for example see that some students do better with AI and others learn less. Students may also ask AI outside of bloomy and I could not find how that is precisely handled. Lastly the course sounds very independent. Are there any group projects or assignments that can be done together. We are for example currently exploring adding tags to assignments to indicate you could pair a programming assignment or that you could pair with AI.

[1]: https://sd42.nl/

These are really thoughtful points. Thanks for sharing them. I'd love to check out your code.

The percentage mastery threshold on Summit may indeed be quite high, but the Summit also comes after the student has proven they are ready for the it by satisfying the BKT threshold on Climb. Still, though, it is worth testing out mastery percentage thresholds to see if lowering it makes any positive difference (e.g., maybe we are not letting students progress fast enough because the bar for quasi-perfection is so high).

On students asking AI outside of Bloomy: that is always likely to be a challenge. One way we mitigate this is by protecting the clipboard so students cannot copy/paste in or out of the platform. This is especially helpful in our Writing Studio tool, where students write long form essays (although clipboard protection is optional and can be disabled by the teacher/parent if they wish -- e.g., if the student typically drafts in Google Docs or Word and wants to easily paste).

Alex, thanks for building this. I imagine it will help many children!

Couple comments on this being 'screen' oriented, which are as you say endemic in schools and terrible.

There's a fair amount of research that paper reading and handling and writing gets significantly higher retention than screen reading. Most especially physically writing notes, but generally comprehension is just higher for paper. Interestingly e-ink is in the middle between screens and paper.

VLMs offer the interesting possibility of 'looking over the shoulder' of students as they work examples and problems by hand through a camera. It's a harder lift, and it's more expensive, but it might find much higher buy-in to think about a combo book / workbook / AI tutor business model, not least because if you talk to almost any parent or teacher they will tell you that they are extremely worried about genz/alpha kids' abilities to focus, read with attention for longer periods of time or handwrite.

Many of the exercises can be evaluated through a VLM pretty much as easily as in a chat window -- more expensive inference, but possibly a much better world.

A middle ground here might be experimenting with something like a remarkable tablet - it's relatively easy to vibe code a responsive eink page right now, and you could do away with the camera side. I guess I'd think of pairing it with a phone app to talk that could hook up to the remarkable.

My pet theory on handwriting notes and why they're so much better than typing or just listening - if not in shorthand, they are too slow to capture lectures word for word, and thus force a first summarization/categorization step, meaning stuff has to literally go in to short term memory then get pulled out and processed; this just has to be better than typing in a sort of fast typing haze and trying to read them later. It's also the reason my college notebooks have things like (WTF?? <---) next to notes -- I was trying to synthesize and couldn't in the moment. In addition to all this, I understand brain scans show quite a high percentage of the brain gets involved when drawing/handwriting notes -- you have micro and macro muscle movements plus all the cognitive and visual tasks together.

At any rate, it will be a great service to learners to have a quality AI tutor, and I'd pitch you on looking seriously at e-ink or paper modalities if you want to help students learn even better and faster. The tech is SUUUPER close right now, or even could be there with a little willpower.

Thanks for the thoughtful comment. We’ve actually had a few school districts ask about paper workbooks, so this is already on our radar. Right now, students can move at their own pace and use a notebook alongside Bloomy, but we haven’t figured out how much paper or e-ink should be part of the product itself.

The VLM idea is especially interesting. It could let Bloomy see how a student worked through a problem, not just whether the final answer was right. We’ll definitely take a closer look at this, so I appreciate your recommendation!

For education, isn't the problem we are trying to solve the one with motivation, values and emotional handling? The tools, play a part, but it is engulfed by the real problem mentioned before. How are you going to deal with it?

When AI can handle the content, teachers get to focus on motivation, values and emotions.

I started my career as a teacher and spent > 8 hrs a day lesson planning, delivering content & grading. If I could have handed that off to a great program like Bloomy, I could have built far better relationships & my students would have been so much better off.

There are many problems to solve within education! To say there’s only 1 aspect would be a massive disservice to the ~2 decades the average person spends going through a school system. Some studies estimate the average American is at a middle school reading level. That’s one problem (of many) we can aim to solve with something like Bloomy.

One metric I would add is mastery under decreasing assistance. A student who answers correctly after three increasingly specific hints has shown something useful, but it is not the same evidence as solving a fresh version unaided. If both outcomes update the skill score equally, the personalized path can advance the student too early.

For each attempt, I would keep the skill, item variant, number and specificity of hints, revisions, elapsed time, and whether the final reasoning came from the student. Advancement could require an unaided success on a new item plus a delayed check later in the session or on another day. The tutor's own explanation should never count as evidence that the learner mastered the step it just supplied.

This also gives teachers and families a more useful explanation than a single mastery percentage. “Can solve independently, but does not retain it after a day” and “can solve with one conceptual hint” point to different next lessons. That distinction seems especially important when the chat interface is both teaching the skill and measuring it.

Great points! Agreed on the distinctions. With enough data we should be able to identify lots of cool things about mastery.

I'm quite surprised by the difference in HN users' (quite positive) reaction to this product compared to quite negative reactions to Ello, posted here a few days ago. Is there a substantial difference in teaching methodology, or other factor that makes Ello unpalatable?

https://news.ycombinator.com/item?id=48852199

Ello is targeted at 5 year olds whereas this is targeted at a broader range. Also the mechanisms are explained much more clearly here.

Childhood psychologists are warning about giving Ai to children, very similar to social media and generally screen time.

How does this factor into your business?

I just posted something pretty relevant to this in a different comment. Totally agree with the dangers. 2 important points:

1. Not all AI is the same. I see 2 MECE categories in education: AI that does all the thinking/execution for you (the outsourced brain), and AI that is designed to fulfill specific coaching/tutoring roles. If we design AI to ask great questions and find the Socratic thread with students, we can actually make students think harder and learn faster.

2. Social media is full of cheap dopamine. See my comment about 'effortful dopamine' for my perspective on this.

The focus on mastery learning via adaptive curriculum is a much-needed approach for K-12, especially since the biggest bottleneck in education is often the one-size-fits-all pace. I am curious how you plan to handle the feedback loop between the AI tutor and the student's emotional engagement with the harder subjects. I am working on something similar for adult personal growth at https://mindsolutions.ai/r/hn , where we use personality assessments to tailor the learning experience, but managing the pedagogical side for younger learners is a much more complex challenge.

curious how the tutor handles a kid who's confidently wrong. from the description the ladder kicks in at the first sign of struggle, which sounds like it's built to prevent mistakes rather than use them. kapur's productive failure work points the other way, students who commit to a wrong approach and then get shown exactly where it breaks tend to keep the concept. the ones who got steered away before making the mistake don't.

does bloomybot ever just let a wrong approach play out and do the postmortem after? also when a struggling student gets rerouted to an easier skill, can you tell that apart in your data from one who was thirty seconds away from the useful kind of failure? the summit gate tells you they arrived. it doesn't tell you whether the help in the middle did the teaching.

Good questions. I like that you're digging into the mechanics.

First principle: BloomyBot does not proactively engage the student unless they make 3 mistakes in a row in "Climb" (the practice portion). Then it engages them.

Otherwise, students can engage BloomyBot when they feel they need assistance.

If the student is confidently wrong and attempts to answer a question, they'll see the feedback that they're wrong, with an explanation. They can then clarify with BloomyBot. The goal here is (as you reference) to cultivate productive struggle. So we DO want the student to try their approach.

On getting rerouted to easier skills: students get 1 chance to attempt and fail a Climb. Enough questions wrong in a row (assuming they're trying) = signal that they either aren't ready to learn the skill or that they didn't pay attention to the instructional portion in Base Camp. We give them the benefit of the doubt and assume the latter. If they fail Climb a second time, they get routed to a prerequisite skill.

The Climb gates Summit through Bayesian Knowledge Tracing. Wilson et al. (2019) derived an “85% rule” from formal models of learning: training is most efficient when learners succeed at about 85% of attempts. So Bloomy uses BKT (Corbett & Anderson, 1995) to update each student’s estimated per-skill mastery after every response, and routes students to skills within their Zone of Proximal Development.

do you think running LLM output through .lowercase() will hide the fact that you are posting AI generated comments?

Congrats man. I've been in edtech for the last 15 years. I was working on something similar (socratic method, 1:1 LLM <-> student, skills based) but joined a company and had to shelve it due to a non-compete.

You've picked one hell of a time to enter the K12 market. 1:1 device backlash, AI tool scrutiny hightening, teacher exhaustion, no recovery in sight for the recent funding cliff.

Your pilot model is the way to go. My favorite tools all grew out of that model. Stay patient, listen to the early users, encourage them to spread the word. Dont' be above grinding it out at a table at some of the smaller regional educaitonal confreneces.

Sounds like you're already doing this but take IXL head on. They are ripe for the picking.

Good luck!

Thank you! It's an interesting time for sure. In the face of all the headwinds, there are also some interesting tailwinds and opportunities:

-Shrinking budgets forcing reevaluation of longstanding contracts, especially those that provide single-subject or limited grade band solutions -A general awareness that, despite the uncertainty around AI, schools need to "figure out this AI thing" -School choice driving demand for mastery-based curricula among alternative educational options

I love a good regional conference.

I want this exact thing, but I dont want the learning to take place on a screen or device. I want the AI in the background, the tool for the parent and teacher to use to drive the child’s curriculum. This seems trivially easy- printers are cheap.

What is your approach to evaluating LLM outputs?

A few things:

First, every BloomyBot response streams through an inline safety classification, and then an independent second-pass audit sweeps every stored message — a deterministic rule layer merged with a separate LLM classifier, taking the highest-severity result. It flags three categories (crisis, distress, inappropriate) across every language we support, and confirmed flags alert the teacher and school. We also constrain what the model is allowed to do: the tutor won't reveal answers, and it's completely absent from Summit assessments — so an LLM output never decides whether a student has mastered a skill. That comes from scored assessment performance against our 90% threshold.

Second, a nightly cron job runs across all BloomyBot interactions to flag erroneous outputs (so far, none) and safety alerts (these happen real-time anyway, but the cron job is a backup).

Third, I sample the interactions manually. I've read most interactions so far (although this will become much harder with more scale).

Finally, we are building structured evals to draw inferences across grade levels, skills, common misconceptions, etc. Ideally we will be able to identify the optimal level of scaffolding to help students achieve mastery, but that will take more data and time.

Great idea and execution but the design language seems like general llm-generated tailwind. I think you guys are missing a huge opportunity to make the design reinforce the product positioning.

The fastest, most surefire way to nuke your child's brain is to give them a screen.

There is certainly room for software to design a curriculum for the parent or teacher to use. We have tried (and failed) to find such a thing for our homeschool curriculum.

But giving the kid a screen? If you're even a midge honest with yourself, a screen will de-educate your child, leading to the opposite of the intended consequence.

My son gets too distracted by any screen (we have had to implement a total screen ban), so screen-based learning doesn't work after a few days. Is there a way to create an analogue paper version? ie with printer and scanner.

Will ChatGPT dream of Bloom's 2 Sigma problem?

https://en.wikipedia.org/wiki/Bloom%27s_2_sigma_problem

Tell us about the test scores increased? Did you find similar results on the PSAT?

If your metrics are good, increasing test scores is the thing thats 100% going yo get schools on board

Congradulations to the launch! When will lower grade come out? And can you speak more about how you determine what "skills" are relevant to teach?

I think product demo walk through would be very helpful on the home page.

Congratulations on the launch!

How do you solve for "meaning?" Is there some implicit expectation that kids arrive with motivation to discover meaning, or some extrinsic rewards?

As AI essentially commoditizes the mechanical aspects of learning, this is the question we find ourselves asking.

I would like to share our journey running a microschool (where our kids go, along with 40 others now) and how we use AI and what we think about its role.

Some useful context, because this really has informed our approach. I am an engineer turned neuroscientist and the co-author of a popular nonfiction book on the mind (and I share this here because the hard effortful self-driven work of crafting this book taught me more about the mind than anything else in my PhD or later). This work, done during the pandemic when learning was coming online, was what pushed us into setting up the school and gave us the courage to work on it from first principles and not rely on any off-the-shelf alternative pedagogy (montessori, waldorf etc). We looked at (and continue to work on) the full-stack of philosophy, pedagogy, process, practice. So the "what should they learn" is also met with "why should they learn this" and also "how should they be learning this"

The *single biggest* learning for us, which we are now able to articulate clearly, is that schools are set up to fail because there is a well-intentioned but misguided focus on measurement (what are they learning). This gravitational pull results in a focus on mechanics (the most easily measurable thing; long division, periodic tables). This results in a narrow focus on concepts (what David Perkins terms elementitis). Most importantly, what gets left out is meaning. Why should I care about this? With a well-intentioned curriculum and an administration pipette we basically leach out agency and meaning.

So we try to solve for meaning. Our biggest worry with AI is that it focuses entirely on mechanics, with very little meaning. Adults and kids are learning to discount AI-content as automated and inauthentic. We derive meaning from connection. AI can help with the mechanics, but only when meaning has been established. Without this clear-eyed understanding and buy-in we basically have chromebooks all over again, which flattened learning to 2D screens and accomplished nothing (if you look at literacy numeracy scores)

Chatbots are great if you start with meaning and motivation. I already care about the why, so I will socratically find my way to the how and what. But if we do not care about it, which really is the biggest malaise (why should a child who has not yet explored the world or literature they like care about foreshadowing or plot mountains?) Sure, the AI could make it easier, but the kids still need to care?

Wrote about our meaning > motivation > mechanics > measurement revelation and the usual unfortunate inversion here https://blog.comini.in/p/schooling-has-a-meaning-crisis-para...

We need many more people approaching education with ideas and experiments, so thank you and good luck!

Happy that this is being tried. I find it weird, sad, unfortunate, etc. that everyone in tech thinks about scientific discovery and drug discovery as AIs boon to society over education.

Every quality of life metric (life expectancy, income, etc) goes up with better education. I guess it’s because most people in tech grew up in neighborhoods with nice schools and went to college so they don’t think about it.

Anyway, great luck.

This strongly resonates with what we are building at ChemioAI [1], although we have deliberately started with a narrower audience: college students taking Organic Chemistry I and II.

We are also approaching Bloom’s 2-sigma challenge as the design goal. The aim is not another chatbot that gives away answers, but a patient, Socratic coach operating within a structured system of diagnosis, guided practice, feedback, productive struggle, and independent mastery.

One reason we chose college O-Chem rather than K–12 is precisely the concern raised throughout this thread about screen time and introducing conversational AI too early. Organic Chemistry is a notoriously difficult gateway course taken by older students who are already studying digitally. It gives us a more bounded, age-appropriate environment in which to test whether AI can strengthen effortful learning rather than replace thinking.

Your separation of the tutor from the underlying curriculum and mastery system is especially important. We have reached a similar conclusion: AI can ask better questions, provide patient scaffolding, and adapt how concepts are explained, but it should not independently define the curriculum, determine mastery, or become the source of truth. We therefore keep curriculum, pedagogical policy, mastery decisions, and safety controls in a deterministic orchestration layer outside the model.

Congratulations on the launch. It is encouraging to see others pursuing the same underlying educational ambition from a different starting point.

[1] chemio.ai

Move fast and break things.

> What evidence or product behavior would you need to trust something like this with a student?

I would need the product to not exist. I don't want this anywhere near a child. I want humans instructing humans, teaching empathy, connection, and - most importantly - learning to learn.

Anything short of that is just building Brave New World. Electroshock the kids if they reach for forbidden knowledge.

I'm sorry to be so negative since I'm sure you worked hard on this but I find all products in this category to be 100% reprehensible.

"Computer-assisted learning. What an insult, to have the computer teach the human."

- Russ Ackoff

There's a place for AI to help people, but given there's no shortage of humans it's difficult to see why we are so egar to replace them. The constant affirming nature of AI coupled with the restrictive nature of the safety layer makes AI both mentally dangerous and politically dangerous.