HomePodcastAI Minds Ep 64
Podcast
AI Minds The Podcast

AI Minds #064 | Griffin Christenson, Founder at Gryfn.AI

AIMinds #064
Griffin Christenson
In this episode, Griffin Christenson shares how Gryfn.AI uses voice and reading tech to democratize dyslexia screening through AI-powered tools. In this episode, Griffin Christenson shares how Gryfn.AI uses voice and reading tech to democratize dyslexia screening through AI-powered tools. 
Subscribe to AIMinds🧠Stay up-to-date with the latest AI Apps and cutting-edge AI news.SubscribeBy submitting this form, you are agreeing to our Privacy Policy.
Share this article

Griffin Christenson, Founder at Gryfn.AI. Gryfn.AI offers AI-powered tools to help educators and parents identify signs of dyslexia in children through a fast, research-backed online screening test. The platform provides detailed insights into reading ability and potential learning difficulties, using speech and language processing to deliver personalized reports and recommendations.

As an optometrist and computer science enthusiast, Griffin Christenson, O.D., is merging clinical insight with tech innovation to reshape how we screen for dyslexia. With a background in biology and a doctorate from Pacific University, he’s developed an AI-powered MVP that digitizes and automates dyslexia risk detection—making early intervention more accessible. Now refining the project through global entrepreneurship training, Griffin’s mission is to fuse vision science with cutting-edge tools to improve care. Outside the clinic, he enjoys time with his wife, dog Lambeau, and cat Oliver.

Listen to the episode on Spotify, Apple Podcast, Podcast addicts, Castbox. You can also watch this episode on YouTube.

In this episode of the AI Minds Podcast, Griffin Christenson, O.D. and founder of Gryfn.AI, shares his journey from optometry school to the world of artificial intelligence, blending his passion for vision science with cutting-edge tech.

Griffin explains how his transition into software engineering and Georgia Tech’s computer science program led to the development of Gryfn AI—a speech-powered dyslexia screening tool that digitizes traditional assessments using Deepgram’s API.

He explores the challenges of building a context-free speech recognition system that can accurately detect reading difficulties and his mission to democratize access to early screening tools.

The conversation touches on Griffin’s vision for open-sourcing Gryfn AI to support broader innovation in screening technologies, while also discussing his long-term interest in wearable devices and combating the rise of myopia.

Griffin also previews his upcoming podcast Italk Tech, which will dive deeper into the intersections of eye care, emerging tech, and social impact—areas where he believes AI can truly make a difference.

Show Notes:

00:00 Georgia Tech's Inclusive Online Master's

04:07 Improving Single-Word Speech Recognition

09:14 Clarify Reading Teletherapy Platform

10:30 Digital Transformation of Dyslexia Screening

13:16 Optometry's Wearable Tech Opportunity

More Quotes from Griffin:

Demetrios:

Welcome to the AI Minds Podcast is a podcast where we explore the companies of tomorrow being built. AI First, I am your host, Demetrios, and this episode, like every other episode, is brought to you by Deepgram. The number one speech to text and text to speech API on the Internet today. Trusted by the world's top conversational AI leaders, enterprises and startups like Spotify, Twilio, NASA and Citibank. We are joined in this episode by Griffin, the founder and creator of Gryfn.AI. How you doing, man?

Griffin Christenson:

I am doing wonderful today, Demetrios. I'm stoked to be here with you and excited to talk about Gryfn AI. I got to put out there that Gryfn isn't flying for me. There's actually another Griffin, Dr. John Griffin, who I'm sure we'll dive into too, but got to start with that. So pleasure to be here.

Demetrios:

You just want to set the scene. You are not as arrogant enough to name your company after your first name.

Griffin Christenson:

That's what I mean. I'm a Griffin, not the Griffin.

Demetrios:

I like it. Well, I appreciate you leading with that. I want to talk a bit about your history because you've done something fascinating where you became a doctor and you said, my life is not difficult enough. So now I'm going to take on another gigantic task and I'm going to learn software engineering. What was the impetus behind that?

Griffin Christenson:

When I found myself, honestly, maybe a tough time to have this become apparent was when I was studying for some of my boards exams as an optometry student. I got caught up in the. The Harvard DD50 computer science course. And I said, this is just absolutely fascinating. So it was really a hyper form of procrastinating that piqued my interest. I saw kind of the development of this AI space coming, and I just couldn't stop wanting to learn. So my wife thought I was a little crazy.

Griffin Christenson:

But I came across a program, omscs, an online master's program through Georgia Tech, which was really an awesome fit for me. They let someone with no computer science background to speak of into the program. After climbing the right hurdles, I had to take a few circuit certificates to make that possible, but it's really a sink or swim program where they open it up to a lot of people and they say, if you can figure this out, we're happy to democratize the access here. And so I'm one class away from finishing my master's in computer science, but really it was just a fascination. I had to learn more and so I did.

Demetrios:

And what do you do during these computer science courses? I think that was the inspiration behind creating Gryfn.

Griffin Christenson:

That it was kind of hand in hand. So I think I had this idea in my head and I just took the longest possible road of getting there, which is, let's start with an academic computer science background. I didn't go to a boot camp. I didn't, just start coding or something like that. This was before a lot of the AI tools too. So I took a very long road. But I think in the back of my mind, my dad's an optometrist and eye doctor as well, and he was an academic before he went into private practice. And he had worked with the Griffin.

Griffin Christenson:

Dr. John Griffin had developed the first diagnostic test for dyslexia. And my dad worked on with him taking that test and making a screening version that could be done in 10 to 15 minutes and detect dyslexia in a much more rapid fashion. Now it's a screener, not a diagnostic test, but in my head, I always thought, that'd be really cool if we could put this in an online format. And like I said, I took a very long path to getting there.

Demetrios:

What is the screening? How can you tell?

Griffin Christenson:

So the screening, it consists of a couple parts. It's basically kind of a fancy spelling test and a fancy reading test. We use someone's reading level to determine what words we want to quiz them on in terms of spelling. And we take that information and compare it to some academic achievements for this grade attained and some information like that. And it lets us detect if someone's at risk for dyslexia.

Griffin Christenson:

And so that's what we've done with Gryfn AI here. And that's where Deepgram became so useful, you can imagine doing a reading test and trying to figure out what someone said into a microphone and so forth is very useful for that test. One of the things that we've run into, though, is different for our use case than a lot of the traditional maybe speech to text use cases, is we're oftentimes using one word at a time. So there's no context for these models. And so that's something where I've really found a lot of utility from the models that you guys offer. They do a heck of a job, but there's some tricks we got to do to make it work appropriately. So maybe, maybe I can inspire some of the folks out there creating models to look at one word at a time would be maybe a goal of mine.

Demetrios:

Because when you say this one word at a time, can you break down exactly how it works?

Griffin Christenson:

So basically, when someone goes through our test, which you can find at Gryfn AI, we will ask them a few intake questions and we get started with the test pretty quickly. When they're presented with the test, they will see a word on the screen. And we have a tutorial that walks folks through this. So they aren't just showing a word immediately. It kind of shows them what to be prepared for, but it will display a word on the screen for a brief period of time, about three seconds. Within that three seconds, they have to say the word that is shown on screen and then the model will detect very binary.

Griffin Christenson:

Either they said that correctly or they didn't is how that aspect of the test works.

Demetrios:

So on the backend, when you're doing this speech to text you sometimes, basically the model is going to either understand or not understand. Is that how you give it the check mark?

Griffin Christenson:

Basically we just do a string comparison. So we'll get a response back from the model and we say, does is this string match what string we wanted? So if the word was cat, did we get cat back? Now if the models. Actually, sometimes it's useful if a model is consistently wrong. If it comes up with the same input or the same output, rather, every time we can account for that. It's where it's subtly wrong in different ways is where it gets to be a little bit trickier. So we've messed with some of the things like that. The key term option you guys have with Nova 3, and then it tends to go the other way where it's too lenient. Now we have a problem where it's marking everything correct.

Griffin Christenson:

And that can be a problem too. So there's a balance there. But it's tough without context for a model. And that said, we're pretty happy with the performance we're getting, but it's going to be an area of incremental improvement. I'm sure.

Demetrios:

It's even tough for me if I say the same word like five times in a row, I forget how to say that word.

Griffin Christenson:

And we have some crazy words, not to spoil it, but like monocotin and lay motif, and it gets pretty challenging towards the end of the Test there. Those are kind of edge cases when someone is getting to that level. But it is interesting to see some of the responses you get back.

Demetrios:

You do these super hard ones because you need to test an adult, I would imagine.

Griffin Christenson:

We test both adults and kids. And again, this is coming from the paper and pencil versions of these tests that already exist. I didn't create anything that way. I simply digitized these and you can have very educated individuals that have pretty extensive vocabularies and are fairly competent at spelling, but it's lower than what we would expect given their attainment. So we will test someone on their reading capability all the way up to postdoctoral level is kind of the last level. But if someone's able to read at that level but their spelling is significantly diminished, that oftentimes aids in a diagnosis. So this is one fact I like to share about dyslexia. And kind of the importance of it is we find folks in the prison and juvenile delinquent societies are overrepresented as having dyslexia.

Griffin Christenson:

We also find CEOs are overrepresented. About 30% of CEOs have dyslexia, which is mind blowing. But it really shows the difference between or the importance of intervention and what the outcome can look like based on the appropriate intervention and so forth.

Demetrios:

Kind of two extremes of society. And you think the intervention is in what way? Like, how do you get nudged towards the CEO path versus towards the delinquency path?

Griffin Christenson:

I think that's exactly what we're trying to be at the center of here. So we're really in the detection space with Gryfn AI. But my wife, it's a family affair with my dad having his background and my wife is actually a speech language pathologist. And so really, what my goal for this detection, because of the importance that I just laid out to you, I want to spread this test far and wide. We're going to open source the code that we have available. I would love for other people to contribute and help improve what we've developed to this point. But we have to sustain it too, at some point. So when we detect someone with these difficulties or at risk for dyslexia, then the next step is how do we help them? And that's really where my wife's focus is going to come in.

Griffin Christenson:

She's working on a platform called Clarify Reading, which we have a landing page coming soon, but she is going to set up a Space where we can do teletherapy and work on some of the skills that can help, nudge people in that CEO or just, whatever somebody's ambition is. But you can imagine the frustration that a lot of people experience with having difficulty with the written sounds and so forth associated with our language and why people end up going that other path. And we're trying to be right at the center. The other thing that's exciting, Demetrios, is a lot of states are starting now to mandate dyslexia screening, and that has a lot of pros. One of the cons, as I see it, is now we got to do more training with these already overburdened school folks. And I think we can really partner with schools and automate the screening process as well. Something I'm excited about.

Demetrios:

The screening and then come in and help them when they need it. That makes a lot of sense. Are there other areas? I really appreciate that you're open source in the code and this whole project. Do you see ways that folks can build on what you've already created to extend to outside of dyslexia? Maybe there are other adjacent disabilities that can be screened for?

Griffin Christenson:

I would say absolutely, yes. Now, I think part of that's just going to be opening eyes. I think a lot of people that maybe have a clinical background similar to mine and maybe not the technical background don't realize some of the ways that we can take a lot of analog type tests that we've had and put them into a digital format. So I think just opening eyes to that is a big part of another thing that I would say is that I think there's other dyslexia screeners that exist too, and there's strengths and weaknesses of these screeners. But, I don't know that those screeners would take a ton from the exact model that we've created. But they could create a similar thing but implement their version of a screener, if that makes sense.

Demetrios:

So in a way, you can now get three or four different tests and have a much better picture or much more confidence if you are. Aren't dyslexic.

Griffin Christenson:

And at the end of the day there's a difference between a screening test and a full blown diagnosis and working with a psychiatrist and going through that process. But I really want to democratize access to this screening and make it available to as many folks as possible.

Demetrios:

Now there's still something that I can't wrap my head around when you were saying all those big words earlier. I do not think I read at a postdoctoral level. Maybe I do. I don't know, throw some of those big words in front of me again and we'll see how good I am. But I also don't think I'm dyslexic. So how does that match? On the test, if I couldn't read those, would you diagnose or would you, screen me as potentially dyslexic?

Griffin Christenson:

It's a good question. There's different levels of reading and spelling competency. And generally what we see with someone who's dyslexic is a disconnect between their attained level of education and where we see that spelling and reading level and so forth tends to be the differentiator. Now, there are exceptions to this because we oftentimes see folks with dyslexia have these great compensatory mechanisms where they'll actually be super strong in other areas besides reading and spelling, and they will get around some of this and achieve these high levels of academic success that makes them stand out on the test more. We see how it's important to define dyslexia, too. We would define it as difficulties with these reading and spelling issues in the absence of an overall intellectual deficiency.

Griffin Christenson:

But very good question.

Demetrios:

What are you looking at trying to conquer next? I mean, you mentioned open sourcing this.

Griffin Christenson:

Demetrios:

Are there any other things that you think are the next frontier?

Griffin Christenson:

There's a couple things that I have on my list. One, I'm just really interested in this space of the intersection between eye care and technology. I think the wearable space with glasses is going to be a really cool thing in the coming years. And I would love optometry to seize an opportunity to be a part of that. We go through the prescribing of glasses and our opticians help us with the fitting of glasses and so forth. And I think there's a little bit of a disconnect with optometry not yet fully embracing this wearable technology. And part of it's because I think the technology needs to get a little bit better for everybody to want to adapt it.

Griffin Christenson:

But I would love us to seize that opportunity. The other things that I have on my radar is I'm actually launching a podcast of my own called Italk Tech, where I'll be hosting people that have a similar interest in this space and the intersection between Eye care and technology. And then the last thing I would point to is you might be a little bit familiar with nearsightedness, but we see this as a growing problem. And I am really excited about some of the future technologies that are coming out to mitigate that problem as well. So those would be some of the big things that I have on my mind as we move forward here.

Demetrios:

Growing problem because we're all staring at our phones.

Griffin Christenson:

That has an impact for sure. We're seeing the rate of nearsa. The percentage of people who are is actually escalating and escalating pretty quickly here. We saw this in the east first. It's now coming to the West. Fortunately, we're developing some interventions that can help mitigate that as well.

Demetrios:

The East? You don't mean New York, you mean like India?

Griffin Christenson:

The Far East? The Chinese population has a particularly high incidence of nearsightedness. And what we're seeing sort of their trend lines that they had there a little bit before, folks in North America. We're starting to match those trend lines here, actually.

Demetrios:

Interestingly now the podcast. Me as a podcast lover, I'm definitely going to check it out. Who are you planning on having on? Because that sounds exciting how you broke it down. Are there folks that are also doing this in this space that are creating other applications?

Griffin Christenson:

I think that's something I'll learn to some extent. I think there is a little bit of a void and a lack of creators in this space is partly what motivated me. But I look to have folks from within my industry, other optometrists. But I really want it to be a more of a far reaching platform than that. So I'm seeking to host founders, I'm seeking to host consumers even. People that have an interest in this space. And so I wanted to really make it for both consumers and clinicians alike and bridge that gap.

Hosted by

Demetrios Brinkmann

Host, AI Minds

Demetrios founded the largest community dealing with producitonizing AI and ML models.
In April 2020, he fell into leading the MLOps community (more than 75k ML practitioners come together to learn and share experiences), which aims to bring clarity around the operational side of Machine Learning and AI. Since diving into the ML/AI world, he has become fascinated by Voice AI agents and is exploring the technical challenges that come with creating them.

Griffin Christenson

Guest

Check out these Resources

View All