I Accidentally Started a Small Business Three Weeks Ago

My Happiest, Most Grounded Distraction

Three weeks ago, I accidentally found product market fit. I’ll phrase it that way because saying it in a more earnest and sincere way makes me too emotional. So, anyway, I built a communication app for non-verbal children that makes 100% of mothers with non-verbal children start sobbing uncontrollably. My son and I discovered this the first day we brought it to speech therapy and some moms saw it in the waiting room. Then it made his speech therapist sob for like five minutes straight, which ironically made her unable to speak. All of this also made me stare at the ceiling and swallow several times because I don’t think it’s good for my son to see me get emotional about his autism.

The thing is: It worked. Or, at least, it worked much better for my son, in particular, than anything else has ever done. And when all the moms in the waiting room saw that it worked, I realized, and I promise I hadn’t thought of it until just that moment, there was absolutely no way I could just shrug and say that my son is the only kid that gets one. I have way too much going on, but I had to make the time to do this specific thing even if it meant less sleep for several weeks.

I need to make several digressions at this point, but I promise I’ll get back to the happy/sappy Hallmark part of this shortly.

What Happens when Your Kid Can’t Talk

Firstly, it’s very unclear your child has speech problems because all children are born without the ability to speak. A year to two years will pass with you none the wiser. Then you and your wife will notice your child should be talking and isn’t, and the pediatrician will say it’s fine. Then you keep on noticing and your pediatrician starts to go over diagnostic questions with you and maybe frowns a little bit. Then you take some behavioral analysis surveys, and you convince yourself that actually you were just being overly dour and it’s fine because you filled the questions out wrong. It’s really bad evidence, you’ll say. Then you look into how clustered this data is at early stages of development and you laugh that it’s going to be totally fine.

One day you’ll see your kid around other kids their same age. All those kids will be having full on back and forth conversations, and you’ll accept that something is actually not fine.

There is a whole rabbit hole of what to do about it. A bunch of would-be wizards will show up to try to sell you things that are definitely not real. Your child, they will say, really just needs to see a pediatric chiropractor. And not even a battle-hardened chiropractor who works on loggers and has just learned to be a very good masseuse, like the one I see, but some piece of shit who cracks the necks of toddlers because he thinks he’s a spiritual healer from another realm. You’ll tell the person telling you to go to a pediatric chiropractor that you’re not interested, but sigh and accept that your wife is going to give your child a lot of “supplements” that you will then verify aren’t poisonous but also probably don’t do anything. Some of them will actually do something, or at least seem to, but they’ll also be the more standard stuff people take. Like iron, or ferritin, or some actual nutrient not in their normal diet.

Eventually, after you go knock on all the right doors, you go to a speech therapist. This is actually helpful, but the speech therapist will really only be able to do so much to help because a lot of this is dependent on your child’s natural ability. This is still worth doing because the speech therapist will know all kinds of techniques you don’t know. If your child is deaf, or mute, or something like that then the speech therapist will have you covered. If they have some sort of body problem, there are all kinds of solutions to help them communicate. If there’s a cognition problem, well… I’m sorry. You’ll do a lot of drills and hopefully you’ll see some progress but that’s about all there is to help you.

I’m not Slamming AAC Devices, They Just Weren’t What We Needed

One of the aid devices they’ll give to your child if they’re struggling to speak is called an AAC, which stands for Augmentative and Alternative Communication. Really, it’s just a tablet full of words and symbols. That’s all it is. Barely one step above a file directory. You combine the words and symbols together to make sentences. The device says those words out loud. This is really good if a child just can’t physically make words, but this is primarily a device for people who are physically impaired but understand language just fine. It’s not built specifically for children and it’s definitely not optimized for someone who is struggling to understand language in the first place. When I looked into the history of these devices they were first built for paralyzed adults who needed to select things based on gaze, and looking at the layout of all of them, that makes perfect sense.

We tried to use an AAC for about a year and my son just wasn’t having it. He’d play with it for a few minutes and then just put it down. Onto the next toy. He only had one at speech therapy and school. I didn’t prioritize getting him one at home because he didn’t respond to it. The whole cadre of therapists and teachers agreed he wasn’t interested in the tool and couldn’t be made to be interested. Instead, we did drills. Interrupt what he’s doing, try to get him to say a word to direct actions, etc. He loves to wrestle and when he speaks spontaneously he most often does it for things like “Again” when I throw him on the couch. We are relatively lucky, because he can talk even if it’s something like one percent as often as other children. Some kids don’t have any words at all.

Everyone kept pushing on me to get him an AAC who doesn’t work with him directly, and I was so busy and so frustrated at that point, I didn’t really want to check a box just to check a box so I told my wife that I’m just going to make him one. I had the idea for a while because it wasn’t super hard to see what he needed. The problem with the AAC is the symbol system. What it does is what he struggles with. If this is your first time reading me, I guess I’m a Forward Deployed AI engineer? There are people who work for me who describe themselves as AI engineers, at least. I don’t really think of myself in those terms, but that’s probably close enough for your understanding.

A traditional AAC was way too abstract. My son has no clue that a red octagon means stop. He has no idea what stop even means in that context. Likewise for the arrows or the other really small stick figures. So I sat down, vibe-coded a cheap website in two hours with some basic navigation and made a few hundred words of vocabulary in chatgpt images where all the images looked like him doing various things, or his actual stuff. And I did all of this in his favorite animation style from his favorite shows. I cloned my voice because I’m the person he listens to the most often, and I made it so that when he pressed the buttons they all said the word or phrase in my voice. I laid out the words in what I figured would be a way that makes sense to him.

We had touch screen laptop so I just loaded the website on there and showed it to him.

What I Made

He was immediately obsessed. This wasn’t just a symbol system to him. It was his own actual life, like a story book or one of his favorite cartoons, laid out in front of him. He recognized himself in all of the images without effort. There wasn’t a picture of a cheese bagel, there was a picture of his cheese bagel. Not a generic picture of a toy, but a specific picture of his toy. He kept pressing the picture of my father over and over again. Then he said the longest sentence he has ever said, which was: “I really love you a lot.”

My dad says this to him all the time.

There was a lot of looking at the ceiling and swallowing at that one.

We took pictures of all his food and explained that the picture with his hand outstretched means “I want” and showed him how to get to the food list. For four years, I haven’t been able to figure out what this boy wants to eat and then suddenly it was trivial. “I want orange” “I want peanut butter crackers” “I want French toast sticks.” What toy has him so upset? Click the picture and it’s just done. These are such small-seeming but major victories it made me legitimately pissed at myself for not doing it before. We have laminated flash cards we even have photo albums, but nothing prior to this that was made specifically to catch his attention. In product terms, I built the product around him. He needed to be able to tell me to smoosh my head against his head, so I made a tile he could press that says, “I need head pressure.”

Then I got mad about the whole state of the field. I promise there’s lots of joy at progress, as well, but then a lot of anger. Probably all misplaced as people who work with disabled children are all sweethearts, but I found myself getting grumpy. You know these things don’t even come with metrics? Well one does, but it’s almost no information. You can’t even see if your kid is getting better over time. Or how often they press which buttons or when. And when you ask if any of them have a teaching mode, nope. Not a single one. Again, they’re for people who basically already know how to talk. And you find fights over things that help nobody, like about how vocabulary shouldn’t be gated off based on ability, and then you get extremely mad because a mom who is exhausted trying to get her child to eat would love to a have a menu that would just solve that problem right away especially if that’s as far as the child’s language ever develops.

So, I built in metrics and a teaching mode and over two weeks his ability to recognize and spot a tile from a word more than doubled. Then I put it on a schedule so he hears a song play and goes over and plays his matching game or listens to it give him a slide show. It asks him every forty five minutes if he needs to go to the bathroom. When we drive to appointments I tap a few buttons on my cell phone and his tablet goes into teaching mode about stuff he’ll interact with at that appointment. If I want to ask him something and he’s not getting it, I type it my phone app and it appears on his screen using the symbols he’s familiar with like hieroglyphs. Suddenly he understands, because the place where he now funnels his attention has surfaced the meaning to him.

After three weeks of this, he is an unhappy little boy. I mean, pissed. He’s super frustrated because he now has a voice and he’s not used to having a voice. Understanding has come with responsibilities. Suddenly his hungry or grumpy vibes aren’t good enough and his mother and I won’t spend time guessing with him and instead point at his tablet and repeatedly say “Tap to Talk!”

He is talking, by the way. Much more than before. Maybe five times as much, so way more than his baseline. Not only will he press the button for pretzels but add in, “I like that” with his own voice. Not only will he press orange but add in that an orange is “yummy.” Last week at the hardware store the cashier gave him a piece of candy and he said “thank you so much” with his own voice.

What I’m Doing Now

My job, as I see it, is to not get emotional about this and keep pushing in the direction of goodness. If there’s anything I can make this thing do that would help some other parent out there, I will make it do that thing. I can’t be the only parent who would really benefit from having a song and slide show playing for their kid about the benefit of eating more foods. We made a few very quick things like this for my son about how it would be great if he didn’t just eat the same things over and over again. So, in the app, he suddenly hears his dad’s voice introducing him to a song that’s all about eating other foods while he looks at pictures of himself trying out other foods.

My son ate a waffle last week. I felt like Rocky at the end of the first movie when he realizes he’s beat Apollo Creed but can’t quite believe it. I have tried everything to get this child to eat more foods including “laying down the law” which has never worked and has only ever approximated trying to give medicine to a cat.

Now I’m figuring out what else to build and how to make this work for other families and I’m going as fast as I can with the limited time I have. To make this actually work across multiple kids and families, for instance, I needed to create a taxonomy of the most common words and phrases that a child acquires over time. Complicated story here but basically I needed a scaffolding I could put into context so when someone is just snapping pictures around their house I can file them into a consistent placement so that all the automation will still work. The parent experience should be they tell me their child’s favorite color and click on an animation style their child likes, maybe they read a script for 30 seconds, and upload a few photos, and then the board just fills itself out. It’s a bit of a headache, but I’m taking the complexity on myself so that when an exhausted mom with four kids hanging off her downloads this app, she has the mental energy to set this up in under ten minutes. Likewise for the routine stuff. I want a parent to be able to upload a couple photos and then I start collecting data and steer the ship from there. This has required me to figure out workable answers to questions like “what does mastery of a topic look like quantitatively?” and “how does that change across different ability cohorts?” For instance, my son has low joint attention. If I have his attention in a given session, he will answer all or most of the questions correctly. If I don’t have his attention, he will press buttons randomly until the test stops. This would take me too long for launch, like a month, but I’ll eventually build in the thing I’m building at my day job —we are in the last stretch before production go live, so that’s why I’m so busy— where a parent can just chat with it naturally and I can start having it adjust the user interface in real time.

His speech therapy clinic wants to use it for other kids, as does his school. I made zero solicitations, so that’s what I meant about accidentally starting a small business. I made it specifically for my son and then, obviously in retrospect, every single other parent with a child like mine, or professional who works with a child like mine, wanted the same thing. There’s lots of paper shuffling to do on my end, for instance I can’t call it a medical device or an AAC, and will only be able to put it in a marketplace with a straightforward description of what it does.

The AAC devices I described above can be expensive —if you get your own hardware directly from the supplier, it can be over seven thousand dollars, but if you have an iPad already you can get a subscription for a reasonable price— but I think I can make this app for $9.99 a month and still do all the AI API connectivity I need to make it as powerful for other families as it is for mine. $19.99 a month if you need voice cloning as it’s more expensive to keep that going. My goal is to make it free when families no longer need to keep generating new images, or at least have some sort of bare substance tier because once you’ve generated a few thousand images you don’t really need to do much more. That’s as cheap as I can responsibly make it because once it’s live I have to keep it on for every kid who uses it.

A few more weeks and then I think I’ll be able to have other families try it out. Anyway, something that I hope is a happy update for many of you.