Saving lives with AI | Emi Gal from Ezra

Thank you.

Right.

Yeah, so we announced last week that we raised $21 million, co-led by Rick at FirstMark and Amir Dan Rubin from Healthier Capital. Before Healthier Capital, Amir was the CEO of One Medical. He's kind of one of the most accomplished entrepreneurs in the U.S., and he joined the board and he's going to help us scale Ezra. And then we also added a number of really exciting investors to the round. We added Allianz Insurance Group, or the Allianz Life Ventures fund, which is part of the Allianz Insurance Group.

We added the Schwarzman family of Blackstone and the former head of the NHS in the UK, Lord David Prior. So we kind of brought in a really exciting group of people to help take Ezra to the next level.

Yeah. So I've been an entrepreneur my whole life. I studied applied mathematics and computer science in Romania, which is where I'm from. I started a company in a very different space than Ezra before, in the advertising technology space, actually building some very nascent machine learning for delivering video ads online. That company did very well, was acquired. I moved to the US as part of the acquisition.

It was in Romania that I started. I moved to London. I lived in London for seven years building it, and then it was acquired by Teads out of New York City, which is a big ad tech company. Teads is actually a French company started by French entrepreneurs. And then I moved to New York and started Ezra because I'm personally at high risk for cancer, for melanoma. I have over 250 moles on my body, which puts me in the highest risk bracket for skin cancer.

And I've had cancer in my family. Sadly, my mother passed away from cancer because she found cancer late. And so I have a very personal motivation to help people find cancer early. And when I started looking at the problem, it came as a complete surprise to me that early detection is really what makes a difference. But the problem is you can't find cancer early in most organs in the body. And so that was kind of the reason why I started Ezra.

Yeah, so I was actually still at Teads and was kind of struggling with this problem, both personally and with my family, and just thinking about it. And I started reading research papers.

And actually, the funny story is the idea for Ezra came on my honeymoon. I was reading research papers on my honeymoon, as you do, and I will never forget a moment. I was reading a paper that was comparing MRI with low-dose chest CT, ultrasound, and other imaging modalities, PET, in terms of their sensitivity and specificity, which is kind of a measure for accuracy in medical imaging. And this paper was concluding overwhelmingly that MRI is the best imaging modality. So I turned to my wife.

We were on these sunbeds on the beach. I was like, hey, if you could do a full-body MRI in an hour, like right now, would you do it to screen for cancer everywhere in the body? And she said yes. And that's how Ezra was born. And then I spent maybe like a year or so trying to figure out what the entry point would be. How would we go about actually building a company from this observation that MRI is a great modality?

Realized very quickly that whilst MRI is great, the problem is it's expensive as a modality because it's really slow, and we can get into that. And with my kind of computer science hat on, I was like, well, I think we can use AI and technology to make MRIs fast and affordable. And so went on the journey and raised the seed round. This was 2018. Started the company in 2017, raised the seed round in '18.

I did, yeah. So I had a computer science and AI background, but I didn't have any MRI background. And so I found a co-founder who had an MRI background, who also had an AI background. And together we built the first AI for Ezra. Even before raising money, we built an AI that was able to identify lesions in the prostate accurately. And then with that prototype, we went and raised a seed round in 2018 that was led by Accomplice in Boston and a number of great investors.

And then we went public. We launched the product in 2019, initially our prostate scan, and then our full-body scan in August of 2019.

Yeah, so it took a while. It took two years from having the idea to actually having a product in market. And there were two things that we did that I think were, with hindsight, really smart things. The first thing we did is we built a large advisory board. We brought on board 22 scientific advisors before we even launched. And these were heads of body MRI at Memorial Sloan Kettering Cancer Center, Siddhartha Mukherjee, a Pulitzer Prize-winning oncologist at Columbia, the chair of oncology at Columbia, some other imaging experts, and we built the AIs and the scanning protocols together with these individuals.

The second thing we did is we ran a lot of tests, validation tests, prior to launching Ezra with people that we knew had issues, potential cancer and so on, that we scanned in order to ensure that we could find the lesions. And we did. And then we launched the product, and it was an immediate success. Both our prostate scan, which was originally prostate MRI, as well as our full body. Our first prostate MRI that we ever did, we found prostate cancer.

And it was a gentleman who actually was from Germany. He happened to be in New York. He saw this thing, he got a scan, and we found prostate cancer for him. And he sent me a heartfelt email a couple of months later saying that we likely saved his life. And this is the first-ever person to get scanned. So I kind of knew early on that we were onto something. It then took a while to get to scale and maturity.

And actually, it was not until last year, so four years in, that we saw escape velocity and started getting—I felt like product-market fit was pulling us into great growth, as opposed to trying to extract growth and trying to kind of convince people to do this scan that we had invented.

So it was a combination of things. One, we had been doing it for a few years, and we had data to show that it works. I met with hundreds, thousands of physicians over the years, just constantly pitching it and showing how it works and why we think it should be something that everyone should get, and so on. And then two, last year, a lot of influencers started getting scanned, like some of them celebrities, famous people, and posting about it. And that got the market to be interested, both on the media side and people wanting to get scanned.

And in the second half of last year, we saw tremendous growth and decided it's time to raise another round, scale, and bring Ezra up nationally.

Yes. And I think the trick is to be ready. Anyone who would want to start a full-body MRI company right now would find it very difficult because there are a few companies that are doing a great job, and we've solved most problems that there are in this space, and we're growing very fast. So you need to benefit from the tailwinds, but you need to have the product and be there. And sometimes it may take years.

To actually get to the point where the market has caught up with you.

Yeah.

Yeah. So YC, this year—every year they post these kind of 20 ideas that they want startup founders to invest in, to spend time on. And one of the ideas this year was MRI, and making MRI faster, making MRI more accurate, basically doing all of the things that we've been working on at Ezra for five years now.

Yes.

Yes. You go to Ezra.com and you book a scan. We have three types of scans. We have a cancer-focused scan, which is a 30-minute, fastest full-body MRI in the world, focused on finding cancer. We then have a 60-minute scan that is a cancer scan plus musculoskeletal analysis. So it does hips and spine. So it will tell you information about that that's not pertaining to cancer that might be useful for your health. And then we have something called Ezra Full Body Plus, which is a full-body MRI plus a low-dose chest CT for lung cancer screening and heart disease assessment.

And I can kind of dive into that. And so we've designed these different scans in order to cater to different price points. So the 30-minute scan is $1,350, the 60-minute scan is $2,000, and then the Full Body Plus, the most comprehensive one, is $2,500. You go to Ezra.com, you book one of these scans, you visit one of our partner facilities, and we don't own and operate facilities. We partner with existing facilities, buy MRI scanning time from them, and run our own protocols, AIs, software on their magnets.

You get the scan there, and then three days later, you receive a report that's not just the radiology report. It's like a translation of the radiology report, which we also do using AI, and we'll dive into that. And then you can book a video call to speak with one of our doctors to go through your results. And so really what we have solved is a way to detect cancer early in all of the organs in which you could not find cancer early before Ezra.

So, not yet covered by insurance. We're working on it. We do do a number of things to make it more affordable. So you can use HSA/FSA dollars. We have Affirm, so you can pay monthly. And the thesis for Ezra, actually, from day one, it's kind of been a constant, has been: build a scan that's high-end, our $2,000 scan. Use the success of that to build a scan that's kind of more mid-market. That's our $1,350 scan, our 30-minute full body.

Use the success of that to build a $500 scan that more people can afford and that you can then obtain payer reimbursement for. And we're currently at the $1,350 scan, and within two to three years, with a lot more AI, we think we're going to get to a $500, 10-minute full-body MRI.

Both. So we build AI to create efficiency in our workflows in order to be able to decrease our cost base, which then enables us to decrease prices. So we're actually the only company that leverages AI at every step of the screening process. We have an AI called Ezra Flash that enables us to acquire MRI images much faster and then enhance their quality in order for radiologists to be able to read them. We then have an AI called Ezra Assist that assists radiologists when they read MRI images.

It was originally cleared for prostate MRIs, and we're working on training it for the rest of the body. And then, once the radiology report is generated, we have an AI called Ezra Reporter that takes the radiology report and converts it into plain English so that anyone can understand what the report means. And so these AIs help us decrease our cost base, and then we use that to decrease our prices. So over time, we'll build more AIs in order to make the scan faster, make the radiology read faster, automate all of our internal processes.

At the same time, the Ezra pitch is that you should do it every year, which means that if you're a member and you do a scan every year, every time you get it, we'll be able to do it slightly faster because we will focus on the change detection as opposed to just acquiring a scan as if we had never seen you before.

Yeah. So in an MRI, particularly over the next decade, most innovation will come from software. We measure the strength of an MRI in teslas, magnetic fields. So a 3-tesla MRI has 60,000 times the magnetic strength of Earth. And we've kind of touched the limits on the hardware front of what we can or should do in terms of putting humans in it. You can also put humans in a 7-tesla scanner, which is much more powerful, but you start getting a lot of distortions if you do scans outside of the brain because the body moves, the organs move, so it creates all sorts of artifacts.

So 3-tesla scanners are kind of the state-of-the-art hardware to be used on humans. Most of the innovation that will come over the next few years will be applying AI to 3-tesla scanners in order to make MRIs faster. The interesting thing about MRIs is you're using magnetic resonance to create images of the body, which is fascinating because you're essentially using a magnet to create images of the body. The way this works is, when you put a human in a really strong magnetic field, what happens is that we have all these protons in the body.

The body's 70% water. That H2O has a lot of protons in it. Protons act like little magnets. They have a north and a south, and they have a spin in a direction. When you put them in a strong magnetic field, they all align to the magnetic field, literally like little soldiers that are now aligned. The way you image using that concept is, once they're aligned in a homogeneous kind of magnetic field, you introduce an interference that aims to flip these protons generally by 180 degrees.

So you introduce literally a radio frequency in order to flip them 180 degrees. You remove this interference, and the protons relax back to the main magnet. Depending on how quickly or slowly they relax back to the main magnet, you can determine the type of tissue. You can determine whether it's bone or liver or muscle or fat, and you can determine whether there's an area in a tissue that's more dense, which could be indicative of disease, and that will pop up as a brighter area on the MRI machine.

So it's a fascinating concept because you're using this ability to flip protons. However, it has a big disadvantage, which is it's a really noisy environment. It's a magnetic resonance environment; it's inherently noisy. So how do you handle that? You do the scan multiple times because the signal will always be the same. The noise will be random. So by doing it multiple times, you can average out the noise and get a higher-quality image. But if you do the scan multiple times, it adds time to the scan.

So what we do at Ezra, or one of the things we do, is we are able to do the scan fewer times, which results in a noisier image. But we have trained AIs to essentially learn what noise looks like in an MRI machine and just erase the noise. And so we've put some images online of the scan prior to the Ezra Flash enhancement and after the Ezra Flash enhancement. It's like they're worlds apart. So that's how we plan to continue using AI to make MRIs much, much faster.

Just going back to your original question of—yeah, fascinating.

Yeah, so the architecture is actually very straightforward. It's a U-Net for Ezra Flash.

Yeah. So we have three AIs: Ezra Flash, the prostate AI, the Assist, we call it now, and then Ezra Reporter, each using different neural architectures.

For Flash, it's computer vision. It's essentially a U-Net, it's called. It's a type of convolutional neural net that has a very particular type of structure that facilitates the ability to train on not a huge amount of data. So a U-Net will have an encoder, a decoder, and some skip layers. Without getting too technical, that will enable you to learn much faster, which means that you don't need to give it hundreds of millions of images. You can give it hundreds of thousands of images, and it will learn how to generalize towards your loss function, whatever that might be.

In our case, we're trying to teach the network what an MRI looks like without noise, and so that at the end of the kind of pass through the network, it knows how to remove that noise. And so it's a relatively plain-vanilla architecture for this network. It's a U-Net. The key advantage for us comes from the data. We have thousands of people that we've scanned, which results in millions of MRI slices that we can use to train. A lot of this data is longitudinal because people come back and get a scan every year.

And for a lot of the data, we have the follow-up downstream data. So if we found something, was it indeed cancer? And we follow up with folks and so on, and we find out whether it was cancer. So it's like a really complete dataset. And without a dataset like ours, you can't actually build a lot of the things that we build.

Because it's a really unique type of thing. When you go to a hospital to get an MRI scan, you're generally getting an MRI scan for a reason. You broke your knee or something. Therefore, there are no datasets out there with a blend of healthy people and diseased people with longitudinal measures over time. And that dataset, we're very protective of it because it is what's enabling us to build these AIs faster than anyone in our space has been able to.

We have a data network effect, essentially.

So when we started to train the prostate AI, even before we launched Ezra, we used the public dataset available called the ProstateX dataset that the NCI put together to build a prototype to show to investors that we actually know how to build an AI and so on. When we launched our first full-body scan, it was about 75 minutes. It was not yet using any AI, and we were just selling it as a high-quality full-body protocol without AI in order to build the dataset.

To be able to train the AIs. And then we got our first 1,000 full-body scans in the first year or so, and that enabled us to start training. And then we've since obviously grown that many, many-fold.

So when a radiologist reads an MRI scan, they need to, especially if they find something in the scan, they need to run a lot of measurements and annotations and segmentation. So, kind of draw a border around the region of interest. All of those things take a lot of time, and they're doing it kind of imprecisely because radiologists need to read 30, 40, 50 cases a day, sometimes a lot more, which means they don't have a lot of time to read these things.

So what we do is we build assist tools. That's why we kind of are calling it Assist AI, to make radiologists faster and potentially more accurate. So for our Prostate AI specifically, it automatically measures the volume of the prostate gland. It's a really important measure to do accurately because it could be indicative of prostate cancer. It automatically detects lesions in the prostate and measures them, again, kind of for providing accurate measurements for radiologists. And then it creates these kind of 3D models of the organ and the lesion that can be used for biopsy prep, for explaining to the patient the extent of the problem, and so on.

In prostate specifically, and part of why we started with prostate was this: a lot of prostate cancers are not to be operated on. They should just be monitored over time. However, you don't know which ones you should act upon easily. You need to monitor over time and track them very minutely. Using our Assist AI, our Prostate AI, you can actually measure the size of a lesion very accurately over time. And if it kind of accelerates in its progression, you then know that it's a lesion you should do something about.

Yeah. Well, one thing AI can already, in many cases, do a better job at is finding potential cancer in medical imaging. It's kind of well established. We see it internally in our data. Many groups have published research to show that. That said, AI can only do the what part: here's what I have found. Radiologists are much, much better, and I think will be for the foreseeable future, at explaining the why part, which is a really important component to a diagnostic kind of protocol.

I think radiologists will not be replaced by AI. They will be replaced by radiologists using AI. There's still a lot of resistance in the medical community to AI stuff. I think it's changing in large part thanks to generative AI. If I am to fast-forward five to 10 years, we will need more radiologists than we have now because we will need more medical imaging than we do now. However, I think all of those radiologists will be empowered by AI tools on the scanning side, on the clinical diagnostic side, on explaining reports to consumers, and so on.

So I don't see a future in which we're delivering a report to an Ezra member without that report having been passed through a human.

Yeah, so the report AI has actually had a significant improvement in efficiency for us internally at Ezra. The way the Ezra reports work is, we don't want to drop a radiology report on you that has all these technical terms and you don't understand any of it. We want to explain every single finding in detail and tell you what you should do about it. We used to do this manually. Our doctors—we have a team of primary care physicians internally—used to take the radiology report, spend 90 minutes per report to generate an Ezra report, which is on average about seven pages, to describe to the member what each finding means.

We built an AI that is able to do exactly that automatically. And our medical providers went from spending 90 minutes to generate these reports to five minutes to just reviewing them. So it's had a kind of profound impact on our business. People on average would get their reports within five to seven days from their scan. We now deliver it consistently in three to four days. So it's kind of been an incredible efficiency for us, which has led to a significantly better experience for customers.

Not GPT-powered, but it is. So the challenge in GPT and generative AI in healthcare is that it hallucinates, and you don't want to hallucinate a finding. It's a problem.

So we use a fine-tuned open-source model, not for generating text, but for filtering through a predefined large database of handwritten text by a human. So it's kind of like a very fine-tuned Llama model with very strict guardrails, which ensures that whenever we put something in a report, the ground truth is true, and the AI isn't actually allowed or able to provide text that had not been written by a human. And then on top of that, we have a human review. It constrains the domain of possibility to about 10,000 options that we know are clinically sound.

And then, as a further kind of measure, we also have every single report reviewed by one of our doctors before it's delivered to members. And we think that's the correct way to apply gen AI-type things into healthcare because it's such a critical domain.

Yeah. So there's the false positive in MRI, which is something that's kind of important to address. And then there's this false positive in AI-related things. I'll tackle both. So MRI is an incredible modality because it's highly sensitive. Sensitivity is the measure of how good is a test at finding the disease. A test that has 100% sensitivity will just catch everything. Specificity is when you find something, is it the disease you are looking for? Is it specific to that particular disease?

In our case, if we find something, is it cancer or is it something else? And so MRI is highly sensitive, 96%, 97%, 98% sensitivity. It doesn't miss anything. Specificity is probably around 80% to 90%, depending on the organ. And so what we've developed internally to minimize those false positive rates caused by the 80% to 90% specificity is a number of things that allow us to determine what we should follow up on. So, for example, every single Ezra finding gets ranked on a score of 1 to 5.

That gets done by our medical doctors, but it also gets done by the AI. So every report, the report translation that I was talking about, every single finding not only is an explanation of what it means, but it also has a score in the background. And that score determines whether we tell you to follow up on the thing or not. That's one of the things we do to minimize unnecessary follow-up from a full-body scan.

There's then the false positive on the AI front. In Ezra Assist, how do you ensure that you don't bias the radiologist with something? And the answer we have gotten to there is for the assist AI to provide a lesion finding to the radiologist, the radiologist needs to essentially draw a bounding box around that area, and the AI will say, "Nope, didn't find anything here," or, "Yes, I found something here. Here it is." And that helps with bias. It helps minimize the risk of the AI pointing out something, and then the radiologist just focuses on that area and not on other areas in the organ, which might lead to a missed finding.

And then, so that's how we handle false positives there. And on the report side, on generating a report, we minimize false positives by not allowing the AI to go beyond the scope of our predetermined defined domain.

Yeah, so a lot to unpack there. I think there are 3 areas that I think are really important for any entrepreneur looking at starting an AI company now. First area is the team. Second area is the domain and the, let's call it, domain expertise. And then the third area is the data. So, the team. One thing I've learned the hard way, or 2 things I've learned the hard way. One is, you would think that in AI, skills are very transferable.

Someone who's built an AI in NLP could easily learn computer vision, or someone who's built computer vision in a non-medical field can easily learn to do stuff in the medical field. I have learned the hard way that it's not that transferable. You actually need—if you're trying to push the state of the art—you need people who have proven experience in the very specific, narrow scope of problem that you're trying to solve. At Ezra, for example, we are very fortunate to work with Dr.

Dan Sodickson, who is a kind of renowned scientist. He invented parallel imaging in MRI, which is a way to make MRI scans faster. He invented that in the mid-'90s, made him famous. All the MRI machines out there right now use the technology that he invented 20 years ago. And part of why we've been able to ship Ezra Flash is because he brought in a very specific type of knowledge that maybe a handful of people in the world have. And without that knowledge, I'm not actually sure we would have been able to pull it off.

And so on the team front, that was the first learning: just very narrow, specific type of knowledge, and it's not necessarily transferable. The second learning, which is equally hard to learn, is it's very hard to find an academic who is able to ship a product. And as a result of that, it took us a while to ship our first AI to production because we initially took the very academic approach. We had an academic team, we wrote papers. It took a while to build a team that both had the academic kind of skills and background and competencies, but also the people that could actually ship something to production quickly.

That's the team part. On the domain part, on the kind of problem that you're tackling, I think it's important to—and maybe it's related to the team part—it's important to really understand what you're solving for and why. I think a lot of the time founders will go and be like, "Oh, I'm going to apply AI to enterprise software." And my question is, okay, but what's the insight? What's the particular thing that you know that is a secret that nobody else knows that you can apply to this thing?

In our case, we've made it public. It's kind of like, we know that you can use AI and software to get to a 10-minute full-body MRI. And 5 years ago, not a lot of people knew that. Therefore, not a lot of people started a full-body MRI company. And even more importantly, most people didn't believe that it's possible. And I think now people are coming around and being like, okay, Y Combinator is putting a pitch out. It's kind of like, "Hey, we think this is possible."

Go in and build stuff there. But we've been doing this for 5 years now, and it's kind of like it took a while to convince the community that it is possible. The third part is, in healthcare especially, the limiting factor in building AIs is data. If you don't have access to a really unique proprietary type of data, you don't have an edge. It's kind of like all of the networks we use, neural networks across all of our different AI products, are essentially plain-vanilla U-Nets, CNNs, and so on.

LLMs, nothing special about them. What makes us special and unique is the data, and not just the data that we have, but the things that we know about what you can do with the data, which we don't talk about a lot publicly. But we have some, call it kind of trade secrets, that have enabled us to get where we are and will enable us to get where we want to be, which is a $500, 10-minute scan.

You know that quote, Matt? We don't do it because it's easy. We do it because we thought it would be easy.

Honestly, that's how I thought Ezra would be. I thought, like, five years in, I'm like, man, this is hard. Being a second-time founder, had a success under my belt, it was relatively easy to raise the seed round. I thought, oh, we got this. It's going to be a walk in the park. And whilst the mission and the approach to achieve the mission have stayed constant, it's been essentially the same plan from day one. Achieving it has been much harder than I thought it would be.

And so the learning in healthcare is that it will be much harder than the founder thinks it will be. Certainly, Ezra was maybe an order of magnitude harder than I thought it would be, to get to some kind of level of scale and ship these AIs and so on. That's one. Two, everything takes longer because you are constrained by regulatory bodies and having to obtain FDA clearance for the AIs and so on, in our case. And so you need to add in six to nine months of regulatory work to any endeavor because it's going to take time to get that done.

And three, as a result of all of that, it takes longer, it's harder, it costs more. And so I think actually we've been pretty capital efficient and have been able to deliver a lot on just over $40 million that we've raised, of which $20 million were raised recently. But it is costly because you need a regulatory team, you need a quality management team, you need to have all of these processes so that if you get a clearance and the FDA shows up at your door, which they can at any moment, you're prepared and buttoned up and ready to respond to any questions they might have.

So those are the biggest learnings for me in a healthcare system.

I mean, clearly you've had success, but we've been fortunate to be able to raise. So we had a great seed round, which was $4 million. We had a great Series A led by FirstMark, $16 million. And then now we've raised another great round. It is certainly harder to raise in healthcare than it is in other areas, probably because a lot of investors just don't understand healthcare. And so they get a little bit—you pitch them, and you know in the meeting that, one, they don't really understand it, and, two, they don't really want to get into the rhythm of having to get FDA clearances and so on.

They want to steer away from that. And then you have a lot of investors that are not going to engage because they don't want to go into the space. You have a lot of life sciences, biotech investors that are not going to engage if you're a more AI software-focused company because they do a different type of thing. They want pharma and drugs. They don't want software and AI because that's what they know better. And so it narrows the pool to not a lot of them.

And then within that pool, there are not a lot who are good. And so as a founder, you need to play the numbers game and meet with everybody and hope that in that process you find someone that likes you, that you also like. And for the Ezra raise, I had 100 meetings. I pitched 100 VCs and feel incredibly fortunate to have landed my ideal profile. Amir is just phenomenal. And he's built one of the most successful, if not the most successful, consumer healthcare tech businesses in the U.S.

And so I'm excited about working with him and having him on the board.

Yeah. So the first thing I'll say is mental health is upstream of startup success, and physical health is upstream of mental health. So the way I look at my performance is, if I'm incredibly fit and kind of optimizing for physical performance, I will then be able to have great mental performance. And at the end of the day, being a good founder CEO is making one, two, three good decisions a day, which you can only do if you're in a good mental state.

And so I spend a lot of time ensuring that I'm prepared physically and mentally for the job. So how do I do that? I have a longevity protocol where I do blood tests every month, every other month. I do Ezra scans. I do all sorts of battery of tests: DEXA scans, body composition, and so on. Then I use all of that data to inform three things: my workout routine, my diet, and my supplement protocol. And so I actually have a company that builds supplements for me based on my blood tests.

And so they give me these pouches for a day that I take that change every other month depending on my blood results. I do Ezra scans.

Yeah, so it's a company called Fount. I'm actually a very small investor in it.

Fount. F-O-U-N-T Bio. It was started by Andrew Herr, a phenomenal founder who used to be the head of the human performance program for the U.S. military. And he left the government and he started this company that essentially tries to help you enhance human performance using supplements. They have all sorts of programs like that. So you can send them the blood tests, and they will create a personalized supplement protocol for you and ship them to your home, and you can take them.

It's a really neat service. I obviously get Ezra scans. I do DEXA scans two or three times a year. DEXA is a great measure of body composition, and so kind of muscle mass, body fat, and bone density. And then I have a highly regimented, very optimized workout routine. So I do three weightlifting workouts a week and two cardio workouts a week. And I actually posted a blog post recently called "Workout Routine Optimized for Busy People," because I only have five hours a week to work out, not every day.

And so I kind of, over time, fine-tuned this protocol that allows me to get that kind of physical performance that I want, that then enables me to get the mental performance that I want, that then enables me to do a good job as a founder CEO.

Yes. So basically what I do is, on Monday, I do bench press and overhead press. So I kind of do upper body, followed by some pull-ups, chin-ups, abs work. On Wednesday, I'll do squats, deadlifts, and overhead press. On Saturday, I'll do squats, deadlifts, and bench press. So those are my weightlifting sessions. And the squats, deadlifts are essential because they're these big compound movements. And then on Sunday I do a run, and sometimes on Thursday I do a run, and that's basically it.

And I do some mobility work attached to those sessions. So, like, some foam rolling, some stretching, things like that. It's designed to not take more than, like, five to six hours max.

Exactly. All of this stuff, I actually wrote an individual blog post for each of them: the supplements, the mental health, the physical health, the workout routine, the diet, and the supplements I take.

Yeah. So we're in seven cities now: New York, San Francisco, LA, Miami, Irvine, Vegas, and New Jersey. And we are planning to be in 20 cities by the end of this year. We're in 22 facilities in the existing cities right now, and by the end of this year, we'll be in about 60 facilities. So we really want to push for having Ezra locations, partner locations, in as many large metro areas in the US as possible. So that's one big goal.

And then the other big goal is continuing to invest in AI. We know how to get to a $500, 10-minute scan. It's just a really hard problem that we need to tackle, and it's going to take us probably about two to three years. And so we're executing towards that, which means hiring a bit more on the AI team, hiring a lot more on the go-to-market expansion team, and then just heads down and building these AIs that we need to build to offer a $500 full-body scan pretty soon.

Amazing. Thanks for having me, Matt. This is great. And I think I was the first one to do it in your new studio.

Thanks for having me.