Latest / The Index Podcast / Greg Lipschitz: The AI That Sees PTSD Where MRIs Can't (Firefly Neuroscience)
Transcript
- 0:00SPEAKER_01: Hey everybody.
- 0:01SPEAKER_01: Welcome to the index.
- 0:02SPEAKER_01: I'm your host, and this is our newest host, my friend, Ann Ahoa Ward.
- 0:07SPEAKER_01: You guys have seen her on the show before, and you've probably noticed that we've added some hosts.
- 0:12SPEAKER_01: The index has always been about people.
- 0:13SPEAKER_01: The reason I started the show is because I believe people are worth knowing, and I want to tell the stories of the people building the future of the internet, but it's never been about me.
- 0:20SPEAKER_01: It's about telling the stories of the people.
- 0:22SPEAKER_01: And the show has grown to the point where like I cannot tell all these stories by myself.
- 0:27SPEAKER_01: I needed other people to help me.
- 0:28SPEAKER_01: And Ann and I have been talking about trying to do something for years.
- 0:30SPEAKER_01: Anna and I are going to kind of spread our wings across all of the future of the internet and future of technology.
- 0:35SPEAKER_01: I'm so excited to have you on.
- 0:37SPEAKER_03: Let's get to it.
- 0:47SPEAKER_01: One quick word from our sponsor, Triton One.
- 0:49SPEAKER_01: If you're building on Solana, your app is only as fast and as reliable as the infrastructure underneath it.
- 0:56SPEAKER_01: At Triton, we built the open source read layer for Solani ourselves.
- 1:01SPEAKER_01: The streaming engine that pushes data the moment it hits the chain, the transaction pipeline that lands your trades when everyone else is timing out, all of it running on bare metal in over 20 data centers worldwide, served by the team that wrote it.
- 1:14SPEAKER_01: We power some of the biggest names on Solana, but with Pay As You Go, anyone can start.
- 1:18SPEAKER_01: Sign up, get an endpoint, pay only for what you use, head to Triton.1, Serious Infrastructure for Serious Builders.
- 1:25SPEAKER_01: I've worked with Triton for years at this point.
- 1:27SPEAKER_01: They've been around since the very beginning of Solana.
- 1:29SPEAKER_01: They've contributed open source software.
- 1:31SPEAKER_01: That's the core of the infrastructure of Solana that makes everything work since day one.
- 1:35SPEAKER_01: I couldn't recommend working with a better team.
- 1:37SPEAKER_01: Triton.1 is your answer for infrastructure on Solana.
- 1:40SPEAKER_01: And can you just share a little bit about your background and kind of where you're coming from?
- 1:44SPEAKER_03: I started my career as a developer.
- 1:47SPEAKER_03: I migrated into the field now known as SEO as it was forming through my love of analytics and my time at Apple.
- 1:54SPEAKER_03: I was a developer designer for Apple Enterprise Education.
- 1:57SPEAKER_03: And that's where I really realized that the party for me was going to be on what was the front of the page, not the back.
- 2:03SPEAKER_03: And so I became a technical SEO and have spent my career growing startups.
- 2:08SPEAKER_03: I've worked with a lot of huge startups.
- 2:10SPEAKER_03: I've worked with small ones and made them huge, helping grow them using data.
- 2:14SPEAKER_03: But I also am a tech pundit and go on TV.
- 2:18SPEAKER_03: I've been on TV over a hundred times.
- 2:19SPEAKER_03: And so I'm really excited to be able to do that here.
- 2:22SPEAKER_01: For some episodes, we're going to be doing a little bit of pre-roll where Ann and I just kind of talk shop about who's coming on.
- 2:30SPEAKER_01: The show I want to talk about that's coming up that you're about to watch is about Firefly neuroscience.
- 2:35SPEAKER_01: And I first met the CEO, Greg Lipschitz, about a year ago during the summer and learned about fireflies.
- 2:42SPEAKER_01: And they built a proprietary database of more than 200,000 EEG and ERP sessions, and they turned that raw brain data into a clinical decision supportive tool that's used on conditions like concussions, cognitive decline, depression, ADHD.
- 2:56SPEAKER_01: And uh today, like on the show, we're gonna dig into that.
- 2:59SPEAKER_01: But I just wanted to have a conversation with you, Ann, about this.
- 3:01SPEAKER_01: Super exciting stuff.
- 3:02SPEAKER_01: What do you think?
- 3:03SPEAKER_03: I think this is a game changer for a lot of reasons, but the biggest being that the way that diagnostics have worked in the brain has been going by how the patient feels.
- 3:14SPEAKER_03: And then if it warrants it, then they scan.
- 3:16SPEAKER_03: So this starts with the data first, which I think is great because it just seems like a really sketchy way to diagnose people.
- 3:25SPEAKER_03: So this is to me the promise of technology coming true.
- 3:29SPEAKER_03: This is this is what the future was supposed to bring to us.
- 3:36SPEAKER_01: Hey everybody, welcome to the index.
- 3:39SPEAKER_01: I'm your host, Alex Gahaya, and today I'm joined by two leaders from Firefly Neuroscience.
- 3:46SPEAKER_01: Uh, really excited to have you guys on.
- 3:48SPEAKER_01: It's Greg Lipschitz, the CEO, and Dave DiCaprio, who's president and COO.
- 3:52SPEAKER_01: They are the team behind Evoke, which is an FDA cleared system that trains AI models on raw EEG and ERP brain signals.
- 4:01SPEAKER_01: That's a mouthful.
- 4:01SPEAKER_01: We're gonna learn more about what that actually means.
- 4:03SPEAKER_01: And it gives clinicians an objective, measurable window into brain function.
- 4:07SPEAKER_01: So it's built on a proprietary database of more than 200,000 EEG and ERP sessions, and it turns raw brain data into clinical decision support tools used across conditions like concussions, cognitive decline, depression, and ADHD.
- 4:21SPEAKER_01: And we're gonna talk all about that today.
- 4:23SPEAKER_01: Thanks for coming on the show.
- 4:24SPEAKER_01: Thanks for having us.
- 4:25SPEAKER_01: We're excited.
- 4:26SPEAKER_01: Yeah, so I appreciate you guys coming on, and this is like very interesting to me as somebody who grew is ADD, ADHD, and like I grew up with dysgraphia.
- 4:35SPEAKER_01: I've had many concussions as a wrestler and you know, grappler most of my life.
- 4:41SPEAKER_01: So there's just like a lot I think that would be really cool to learn from you guys about what you're seeing and what people, how this technology helps people and helps doctors.
- 4:49SPEAKER_01: Before we do that, maybe Greg, can you just tell me?
- 4:51SPEAKER_01: We'll go Greg and then Dave.
- 4:52SPEAKER_01: Same question for both of you.
- 4:53SPEAKER_01: What gets you out of bed in the morning to be working on this product and to be at Firefly Neuroscience?
- 4:59SPEAKER_01: Sure.
- 4:59SPEAKER_01: Yeah, happy to answer that.
- 5:01SPEAKER_02: You know, my experience, and I think it's pretty common with others, is that we all have experienced mental health uh in some way, shape, or form.
- 5:09SPEAKER_02: So whether it's yourself or you know, family members or people friends close to you have kind of suffered from anxiety or depression or dementia.
- 5:18SPEAKER_02: And so I've had a number of people close to me, whether it's um family members suffering from anxiety or friends or you know, a couple people in my life who have uh you know taken their life to depression, and the ability to basically provide this tool that we can actually scan the brain and and measure it and identify what's wrong with it, and then help guide you know what's the best treatment for that person is what gets me out of bed every day.
- 5:49SPEAKER_02: And it's hey, you know, if these people would have had these tools, we could have saved lives.
- 5:54SPEAKER_02: And so every day, you know, we're scanning people's brains or over the 1,500 brains a month and growing.
- 6:02SPEAKER_02: And so, you know, how can we get this technology out to as many people as we can?
- 6:07SPEAKER_02: Because I think it really changes people's lives.
- 6:09SPEAKER_02: What about you, Dave?
- 6:10SPEAKER_00: It's always hard going second from here because I think it I think some businesses you ask people like what gets you up in the morning, and there's like a big explanation.
- 6:19SPEAKER_00: Like, we're trying to unravel the brain and solve mental health.
- 6:22SPEAKER_00: Like it's it's fairly straightforward what gets you excited about doing that every day.
- 6:27SPEAKER_00: Your ability to help people, the ability to make a difference.
- 6:30SPEAKER_00: It's it's massive in this space, and it's really, you know, it is the thing that is left in healthcare.
- 6:37SPEAKER_00: I've been in healthcare and medical for a long time.
- 6:41SPEAKER_00: Um and it is really the big unsolved problem.
- 6:44SPEAKER_00: It used to be people would worry about cholera, infectious diseases back in the day, and you know, then we had antibiotics, and people all of a sudden were living longer.
- 6:54SPEAKER_00: And then it was cardiovascular disease.
- 6:56SPEAKER_00: And we figured out still a lot of people dying from that, but we know the factors.
- 7:01SPEAKER_00: If you really care, you can change your diet, you can exercise, and you can get healthy.
- 7:05SPEAKER_00: Cancer has made tremendous strides in the last 20 years.
- 7:08SPEAKER_00: The thing everybody is scared of, for both them and themselves and their parents, is as I age, what's gonna happen to my brain?
- 7:15SPEAKER_00: Am I still gonna be here?
- 7:16SPEAKER_00: I know I can keep myself physically alive, but what's gonna happen then?
- 7:20SPEAKER_00: Um, it's it's you know the biggest concern there is in health right now.
- 7:25SPEAKER_00: And the biggest letter to move the needle.
- 7:27SPEAKER_01: How did this company start?
- 7:28SPEAKER_01: Like tell me the genesis of all this.
- 7:31SPEAKER_02: It's it's a long story, but um basically, you know, my career has been in finance, and you know, I like to invest in incubate companies.
- 7:40SPEAKER_02: And about five years ago, uh I met the chairman of Firefly, uh Arun Menewat.
- 7:46SPEAKER_02: Um, and Run's had you know three healthcare exits.
- 7:50SPEAKER_02: One was uh a medical imaging company he sold to IBM for a billion dollars.
- 7:55SPEAKER_02: Uh he then sold another company, helped build and sell to Stryker for 700 million.
- 8:00SPEAKER_02: And so uh, you know, he he basically explained to me, he said, look, there's this company and this technology, and these guys have spent you know 10 years and 60 million dollars building the largest database of these brain scans.
- 8:14SPEAKER_02: And and why it took that long is you know, when it comes to AI, it's all about the quality of the data.
- 8:19SPEAKER_02: And so everyone had to go through like a standardized scan, and they needed a critical amount of people from the ages sort of 12 to 85, because now the brain's different for a 12-year-old than a 15 or 30-year-old.
- 8:32SPEAKER_02: So you need this big database.
- 8:34SPEAKER_02: And so we had these brilliant people doing sort of research and data collection on the brain, and then getting that through the FDA.
- 8:41SPEAKER_02: And so when Arun found me, they were just at the point of commercializing the technology.
- 8:46SPEAKER_02: And so uh, you know, I invested, I got involved, and um I shared an office with an MIT uh quant trader, and I said, Hey, I've got this amazing technology, I need the best AI healthcare guy you have.
- 9:01SPEAKER_02: And so he's like, Well, you got to go meet Dave DiCaprio.
- 9:04SPEAKER_02: And so flew down to Austin, you know, met Dave, got Dave to join the board.
- 9:10SPEAKER_02: And basically what we realized was we had all this data and this amazing technology, um, but we really needed a simple hardware, you know, for a clinician to use or for anyone to use to scale it.
- 9:22SPEAKER_02: And so, you know, we were basically we're speaking to a group, you know, Evoke.
- 9:26SPEAKER_02: The founder of Evoke was Dr.
- 9:28SPEAKER_02: Hagerdorn.
- 9:29SPEAKER_02: He actually, he's a neurophysiologist, and he basically was treating veterans coming back from the Middle East with PTSD, and he realized like other technologies couldn't actually capture PTSD on the on the scans and you know, MRIs or PED or CT.
- 9:48SPEAKER_02: So he developed this really easy-to-use hardware to help identify that.
- 9:53SPEAKER_02: And so, you know, when we looked about a year and a half ago, we were speaking of Vogue to acquire them kind of merge the software and the data with the hardware.
- 10:03SPEAKER_02: I basically sent the the hardware to Dave.
- 10:06SPEAKER_02: I said, You're the you're the tech guy, you know, do the due diligence and let us know uh if this is any good.
- 10:12SPEAKER_02: And uh Dave had a dinner party and he scanned the brains of you know all the people there.
- 10:18SPEAKER_02: And then a few days later, uh Dr.
- 10:20SPEAKER_02: Hagerdorn you know read it.
- 10:21SPEAKER_02: He he never met any of these people, and he's just looking at these like EEG signals and these brain reports.
- 10:28SPEAKER_02: You know, it's almost like a David Blaine like magic trick in the sense of like he was looking at these things, he's like, this person must have played football, and this person must have had a traumatic event in their life.
- 10:39SPEAKER_02: It kind of was like four for four, the people, and then at that point, you know, we closed uh the acquisition and and Dave joined full-time about a year and a half ago, and we've just been building since.
- 10:50SPEAKER_01: That's wild.
- 10:51SPEAKER_01: Dave, what is your background?
- 10:52SPEAKER_01: What you're the AI guy, where where did you come before this?
- 10:55SPEAKER_00: Yeah, so I did my undergrad in MIT, and then I started in AI and healthcare in like the early 2000s on the Human Genome Project.
- 11:05SPEAKER_00: So I got an opportunity to go work there when they were first sequencing the human genome, and it was No big deal.
- 11:10SPEAKER_00: Just at the time I didn't realize that, oh, this is the thing you're gonna talk about in 20 years when people ask you.
- 11:17SPEAKER_00: It was pretty amazing to be part of that uh while it was going on.
- 11:20SPEAKER_00: My job was I got the raw genome sequence.
- 11:23SPEAKER_00: So it was like three billion A's, G's, C's, and T's.
- 11:26SPEAKER_00: And we would run every machine learning and AI algorithm known, basically, all at scale against that to try to discover things that were going on.
- 11:35SPEAKER_00: It was a great introduction to AI and machine learning and healthcare.
- 11:39SPEAKER_00: It wasn't AI back then.
- 11:41SPEAKER_00: It was like, it was that we just called it pattern detection back then, but then it got a little more investment money and we called it machine learning, and then it got a bunch more investment money, and now it's called AI.
- 11:51SPEAKER_00: Core math and everything.
- 11:53SPEAKER_00: I started doing there.
- 11:54SPEAKER_00: As soon as I was in healthcare, I got hooked.
- 11:56SPEAKER_00: I've jumped around from working on the genome to doing drug discovery and combinatorial chemistry kinds of things, and then ultimately got into patient data, and then now go into the brain.
- 12:08SPEAKER_01: What a wild story, though, about how you vetted this technology, the hardware piece component.
- 12:13SPEAKER_01: Like you literally just had some friends over and you were like, put this thing on your head.
- 12:16SPEAKER_01: Yeah, yeah.
- 12:17SPEAKER_01: It was like, I think that was your due diligence.
- 12:21SPEAKER_00: I mean, at some point you can you can read the papers, you can understand the tech, but at some point you just have to get your hands on it dirty with things.
- 12:28SPEAKER_00: And so, yeah, I had some people over.
- 12:30SPEAKER_00: I was like, I got this EEG machine, I gotta go figure out, like, you know, it was like, hey, can we actually do this?
- 12:36SPEAKER_00: Can I I'd never done an EEG in my life.
- 12:38SPEAKER_00: I was pretty messy the first time.
- 12:40SPEAKER_00: I have to say, my wife has been very patient.
- 12:42SPEAKER_00: She's done, she's gotten more EEGs than any sort of normal, healthy person ever.
- 12:47SPEAKER_00: Her brain is fully mapped.
- 12:49SPEAKER_00: Yes, yes.
- 12:50SPEAKER_00: And I I mean, I still don't understand my wife, it's I think nobody's hasn't helped.
- 12:54SPEAKER_00: We're not that good.
- 12:55SPEAKER_00: It's not that good in the technology, but we can't help with a bunch of stuff.
- 12:58SPEAKER_00: And yeah, and so I I came at it from a data perspective, and it was I'm super excited to see a big mass of data that was just not easy for anybody to get.
- 13:10SPEAKER_00: I was I coming out of software and a software background, I was really worried a few years ago.
- 13:16SPEAKER_00: Like, how do you, as a small company or an entrepreneur, how do you compete when Epic and Microsoft and Google have all the healthcare data there is to have?
- 13:27SPEAKER_00: Like, so how do you, or it's locked up in the big pharma companies.
- 13:30SPEAKER_00: How can you, as a small company, get it?
- 13:32SPEAKER_00: And then we managed to find this asset that was available that had this wonderful proprietary data set that you know took a decade to build.
- 13:39SPEAKER_01: Yeah, I think it's super interesting.
- 13:41SPEAKER_01: So why did it take a decade?
- 13:42SPEAKER_01: Is it just because it was just like there wasn't enough technology and it just took a long time for them to get better technology, get better scans?
- 13:49SPEAKER_01: I mean, what was the that's a long time?
- 13:51SPEAKER_00: Yeah, yeah.
- 13:52SPEAKER_00: A lot of it is just getting the diversity of people and getting the the environment out.
- 13:57SPEAKER_00: You you you need you need a variety of different brains.
- 14:01SPEAKER_00: So, you know, you can't go, you can't just go around the Google office and get a bunch of like mid-20s or mid-30s uh highly intelligent people.
- 14:11SPEAKER_00: You you want to collect scans from people with all kinds of disorders, trauma, uh traumatic brain injury.
- 14:20SPEAKER_00: So to get a really diverse set of data across a really wide clinical range, just it's hard to get all the people.
- 14:28SPEAKER_00: We have you know 200,000 scans, 160,000 different people in our database, and that that just takes a long time.
- 14:35SPEAKER_01: But uh, you know, Greg, it's making me hearing him say that, when Dave say that, it's making me think about the discovery of this opportunity initially.
- 14:43SPEAKER_01: I forget the guy's name, but this doctor sounds like been trying to get this brain data and he started a company around it.
- 14:52SPEAKER_01: He had had a company already, and that was Fireflies.
- 14:56SPEAKER_01: So how did you see this?
- 14:58SPEAKER_01: You know, like how did you like even come across the opportunity?
- 15:03SPEAKER_02: So, well, what what Dr.
- 15:04SPEAKER_02: Hager Dorn is interesting because you know he was seeing all these veterans coming back from the Middle East with PTSD.
- 15:11SPEAKER_02: And you know, if you do this, the structural you know, scans that we have today, you know, MRI, PET CT, etc., you can't really see PTSD in them.
- 15:21SPEAKER_02: But it's really interesting with EEG, it's actually very easy.
- 15:25SPEAKER_02: Like once once you someone explains it to you, like anybody can learn within five minutes.
- 15:29SPEAKER_02: And really, like we scan the brain uh for five minutes eyes open, just resting, and and five minutes eyes closed, and then there's a 10-minute, we call it event-related potential.
- 15:41SPEAKER_02: But just looking at the difference between eyes open and eyes closed, the brain's normal for someone with PTSD when the eyes are open.
- 15:48SPEAKER_02: When they close their eyes, they their brain goes in a hyperdrive.
- 15:52SPEAKER_02: They they can't, they don't have all the information, so the amygdala starts overfiring.
- 15:58SPEAKER_02: And so you can see like when you compare these screens, like everything looks fine, eyes open.
- 16:03SPEAKER_02: The second the eyes are closed, like you could just see this part of the brain that just lights up that's overreacting.
- 16:08SPEAKER_02: And so he was predominantly using it for that purpose, but the use cases are endless.
- 16:12SPEAKER_01: But I mean, what what like how did you even meet him?
- 16:15SPEAKER_01: And like, was it already a publicly traded company when you talked to him, or was it before that, or or was the that was just a vehicle to get the ac to acquire the data from him?
- 16:25SPEAKER_02: No, so Firefly um was this database um that spent 10 plus years building and you know ran AI through it and got approved through the FDA for what's considered a it's a normal brain for ages 12 to 85.
- 16:41SPEAKER_02: And through that process, you know, we met and understood, you know, it's a small industry of people working on these things.
- 16:46SPEAKER_02: And so uh naturally we kind of knew of of Dave.
- 16:50SPEAKER_02: And Dave's this you know, brilliant neurophysiologist, but we recognize like, hey, we have all this data, you've got this hardware, you're a doctor, we're business people, like this is just like the perfect match.
- 17:02SPEAKER_00: And so, you know, when we did the merger a year and a half ago, it's just been the neurotech space is interesting in that we just call everybody up.
- 17:11SPEAKER_00: Every company we hear of, Greg just calls up the CEO, is like, hey, let's chat, let's get on the phone and talk.
- 17:17SPEAKER_00: Um, and it's interesting because it's a small enough, niche enough industry right now.
- 17:21SPEAKER_00: It's it's growing very, very rapidly, and there's some big name companies in the space.
- 17:26SPEAKER_00: But we spent a lot of time in the early days just trying to talk to everybody we could to figure out what's the lay of the land and who are the right organizations to work with.
- 17:35SPEAKER_01: Yeah, it's just so in interesting to me that that you guys discovered this opportunity because they're uh I think the data so I I recently had the founder of Red Sky Health on my show.
- 17:48SPEAKER_01: I don't know if you guys have heard of that company, but they have a product called Daniel, which is an anagram for denial.
- 17:54SPEAKER_01: I ended up investing in the company after the show because I was so excited about what they're doing.
- 17:58SPEAKER_01: And they um they're like a seed stage series A startup.
- 18:02SPEAKER_01: And what they do is they get medical um like doctors' offices signed up and they fix uh insurance claim denials with it with and and they have like their own models that they've trained on the data, and it's kind of similar in the sense that you know Google can't just get that data, like it's all HIPAA private data.
- 18:20SPEAKER_01: And so they have to sign a contract with the only person who has the data is the is the doctor's office.
- 18:25SPEAKER_01: So you have to get an agreement in place, and they do all performance-based.
- 18:29SPEAKER_01: So like the more insurance claims that they get fixed, that undenied, I don't know, accepted, however the right word is for that, the more money they make.
- 18:38SPEAKER_01: And I just think it's a similar in in a in the sense that the data was kind of hard to access and it was in and it's in the medical space.
- 18:44SPEAKER_01: And I just feel like there are more of these opportunities out there than people would think.
- 18:49SPEAKER_01: And there are probably several in the space that you're working in because just because of the nature of the of the space.
- 18:56SPEAKER_00: Yeah, my my prior company that I founded, Closed Loop AI, we we dealt all with claim healthcare claims data, worked a lot with both hospitals and insurance companies on some of those kinds of issues.
- 19:08SPEAKER_00: Uh, that was fun.
- 19:10SPEAKER_00: We won a million-dollar AI competition with that at that company.
- 19:13SPEAKER_01: Yeah, I remember Greg telling me about that.
- 19:15SPEAKER_01: Tell me, tell, share, tell that story real quick.
- 19:17SPEAKER_00: This is back in 2017.
- 19:19SPEAKER_00: We started that company, and the idea was everybody's got their data.
- 19:22SPEAKER_00: So you have some of these organizations.
- 19:24SPEAKER_00: How do you help them build their own custom machine learning model to be able to optimize this stuff?
- 19:29SPEAKER_00: So we basically built a platform for building models.
- 19:33SPEAKER_00: Our product wasn't the model, it was the machine that builds the model.
- 19:36SPEAKER_00: And then uh after about a year and a half, CMS, which is the federal organization that runs Medicare, they sponsored this million-dollar,$1.6 million X prize, basically, which was like, hey, anybody can sign up, you submit an application, and then we're gonna try to see who is the best at healthcare AI in the country.
- 19:57SPEAKER_00: And we're we're like a 12-person startup at this point, but we're like all in on it, and we're just okay, we're gonna go.
- 20:04SPEAKER_00: So we apply, we we end up, we end up getting in.
- 20:08SPEAKER_00: So there was about 400 companies that applied, and they picked 25 to actually start running the trial.
- 20:15SPEAKER_00: Um, and it was funny, I always remember we were sitting there with our VCs one day and we were all in on winning this thing.
- 20:20SPEAKER_00: We're like, our business plan is win the million-dollar competition, then we show we're the best in the in the world at this, and then we go try to sell to everybody.
- 20:29SPEAKER_00: Uh, and I remember our VCs were like, whoa, guys, like, what happens if you don't win?
- 20:35SPEAKER_00: And I I'm like, who who are the you guys are the venture capitalists here?
- 20:39SPEAKER_00: And you're like, no, like this is the street.
- 20:43SPEAKER_01: We're supposed to risk.
- 20:44SPEAKER_01: We're just taking risk.
- 20:45SPEAKER_00: Yeah, and I remember having this discussion where they're like, no, no, no, no, no, maybe you guys should like make sure we're like, no, we're all in, we're gonna win.
- 20:52SPEAKER_00: And they're like, what happens if you don't win?
- 20:54SPEAKER_00: Like, we'll figure that out if we don't win.
- 20:56SPEAKER_00: But that's that's not the plan.
- 20:58SPEAKER_00: It was two years, and it was supposed to be a year long.
- 21:00SPEAKER_00: It ended up happening in the middle of COVID.
- 21:03SPEAKER_00: We weren't sure it was gonna happen at all.
- 21:04SPEAKER_00: Then they ended up doing it, delaying it.
- 21:06SPEAKER_00: They cut it down to seven.
- 21:08SPEAKER_00: And we had like our competitors were like IBM Watson Health, Merck, the Mayo Clinic, the Cleveland Clinic, Accenture, like everybody else we're competing against is a health fold name, and we're this little company no one's ever heard of it.
- 21:21SPEAKER_00: We ended up winning the whole thing, and so that was pretty exciting to kind of get through there.
- 21:26SPEAKER_00: And it the contest was basically a heads up like Medicare would send you data, and it was funny, they did this thing where it was like, okay, it's 7 a.m.
- 21:35SPEAKER_00: we're gonna send you the decryption key to unlock the data, and then you have 24 hours to like do your predictions and upload them.
- 21:42SPEAKER_00: And so it's kind of this race against time to try to get out of here.
- 21:45SPEAKER_01: Okay, you're like sitting at your desk, like hitting refresh, yeah, just like waiting for that key to come through.
- 21:50SPEAKER_00: Oh, and then it happened during the so in I live in Austin in 2021.
- 21:56SPEAKER_00: There was like this massive ice storm in Texas, like all I remember.
- 22:01SPEAKER_00: Yeah.
- 22:01SPEAKER_00: Yeah.
- 22:02SPEAKER_00: It was during that.
- 22:03SPEAKER_00: So I'm like getting up.
- 22:04SPEAKER_00: We had a well at the time.
- 22:06SPEAKER_00: I've got like a hairdryer like unfreezing my well.
- 22:09SPEAKER_00: I'm like trying, it was crazy.
- 22:11SPEAKER_00: Like we hadn't we're working off borrowed power, and I'm trying to, you know, solve it.
- 22:16SPEAKER_01: Yeah, this is before like agents could do things for you while you sleep.
- 22:19SPEAKER_01: You know, it's not like you had some like AI in the cloud, like doing you know, running job, like you had to actually like code.
- 22:25SPEAKER_00: Yeah, yeah.
- 22:26SPEAKER_00: Yes.
- 22:26SPEAKER_00: This was this was before those days where you had to really type everything yourself.
- 22:30SPEAKER_00: Um, and there are two of the seven teams I think got knocked out right at the end because they submitted the wrong file.
- 22:36SPEAKER_00: Uh wow.
- 22:37SPEAKER_00: Yeah, one of the teams posted a question like, can we replace the file that we submitted?
- 22:42SPEAKER_00: And they were like, no.
- 22:43SPEAKER_00: Nope.
- 22:44SPEAKER_00: Nope.
- 22:44SPEAKER_00: You get one, you know, two-year contest, and at the very end, you submit the wrong thing.
- 22:50SPEAKER_00: Like, I was pretty nervous hitting that button.
- 22:54SPEAKER_01: Don't you just kind of feel terrible for them?
- 22:56SPEAKER_01: Like it's just terrible.
- 22:58SPEAKER_00: I told everybody two years, I told everybody, I was like, the team that wins is not going to be the team that does the smartest thing.
- 23:05SPEAKER_00: It's the team whose dumbest mistake is the least dumb.
- 23:09SPEAKER_00: Like, yeah, we gotta make sure that we don't like our dumbest mistake is smarter than everybody else's.
- 23:15SPEAKER_01: Well, let me ask you this.
- 23:16SPEAKER_01: I mean, how's that playing into your role now and what you're doing over at Fireflies and the job, you know, what you guys are executing?
- 23:23SPEAKER_01: How do you does that lesson carry through to your day-to-day?
- 23:27SPEAKER_00: Yeah, I mean, I think a lot of you know, one of the things we did in that contest was you just don't know what's gonna work.
- 23:34SPEAKER_00: You can have a lot of ideas.
- 23:35SPEAKER_00: We had like 280 ideas.
- 23:37SPEAKER_00: We made this giant list.
- 23:39SPEAKER_00: And I was just like, it's how fast can we cross off these ideas?
- 23:42SPEAKER_00: How many of them can we try?
- 23:44SPEAKER_00: And I think that holds up tremendously in the brain space because so much of what we're learning is new.
- 23:51SPEAKER_00: This is like new techniques, there's new data sets.
- 23:54SPEAKER_00: And so we're really trying to figure out how do we go fast and experiment really quickly to understand what's going on.
- 24:01SPEAKER_00: And it's just try the next thing, try the next thing, try the next thing as fast as you can, and you know, double down on what works and just throw away what doesn't.
- 24:10SPEAKER_00: Because you don't know.
- 24:11SPEAKER_00: There's no way to predict what's too new.
- 24:13SPEAKER_01: Not to get overly personal, but you guys are inspiring me on an area where I think I would like to do a project at some point, and maybe it's something we could do together someday.
- 24:22SPEAKER_01: The pregnancy space, I think, is a similar space where there's almost no data.
- 24:29SPEAKER_01: And I I got involved in a uh project through Mount Sinide and another company called Evidation Health.
- 24:36SPEAKER_01: We were the software vendor, we built a bunch of open source software for like a first of its kind study on pregnancy.
- 24:43SPEAKER_01: And the reason we yeah, so so my my company, not my uh another company of Brian and I.
- 24:50SPEAKER_01: So my partner Brian is the guy who wrote Bash back in the 80s um and and like uh did GPL licenses, like OG open source guy, and we became friends with a guy named Luca Fuccini who runs um Sage Bio Networks.
- 25:02SPEAKER_01: You can look that up.
- 25:03SPEAKER_01: That's yeah, so he's um which is a big NGO for biomedical research data.
- 25:08SPEAKER_01: And he and I like, I don't know how we connected, we connected uh through friends, we were grabbing wine one day, and we had just lost our second child, like super late term in a pregnancy, and it was maybe like three or four months out, and so still pretty raw.
- 25:23SPEAKER_01: And he was talking about this study, and I was like, I want to be a part of that, I want to help.
- 25:27SPEAKER_01: And so we ended up building all the application stack that like helped gather that data.
- 25:31SPEAKER_01: It was like a thousand women, but it was around the same time that you were at closed loop because literally right when we were about to launch the study, we had finished the app, it was ready to go, COVID hit, and it like delayed everything about a year.
- 25:43SPEAKER_01: And I I was just talking to my friend Mackenzie, who at the time was the project lead at Evidation on that, was working with us to build the app, and it ended up the study actually ended up happening.
- 25:52SPEAKER_01: I I still haven't read the results, but I think there's something here that's similar to what you guys have done, not with the imaging technology, but just generally speaking, like AI and the data that's not available, that would be so massively impactful.
- 26:05SPEAKER_00: We learn from an astonishingly small portion of the data we collect today.
- 26:11SPEAKER_00: That it it is you know it's crazy.
- 26:16SPEAKER_00: If you read cancer trials, you read so many things.
- 26:18SPEAKER_00: If you actually dig into the research on so many of the medical guidelines that we use today, you find they're based on like 200 people.
- 26:27SPEAKER_00: Somebody did a study 20 years ago on with like 200 people, and then that set a guideline that has now lasted for two decades.
- 26:34SPEAKER_00: And meanwhile, maybe 10 million people have gone through that protocol.
- 26:37SPEAKER_00: And have we measured all of them and do we understand the outcomes?
- 26:40SPEAKER_01: Like, no, it's it's uh yeah, it's it's just amazing to me that there are so many areas like this that are uncharted still.
- 26:48SPEAKER_01: They're like black holes of just not a lot of data.
- 26:51SPEAKER_01: I'm curious, are you guys are you using your own models?
- 26:55SPEAKER_01: Or are you training are you using frontier lab models?
- 26:57SPEAKER_01: Are you training, you know, using open source stuff?
- 27:00SPEAKER_01: Like, how are you leveraging the data?
- 27:02SPEAKER_01: Like what kind of AI are you actually using?
- 27:04SPEAKER_01: Yeah, you can share.
- 27:06SPEAKER_00: Yeah, sort of uh all of the above.
- 27:08SPEAKER_00: I think there's a there is, I mean, we have the data set.
- 27:13SPEAKER_00: I come a little bit old school, like we have a proprietary data set, and we can build some of our own models on top of that for sort of that core understanding of the EEG data.
- 27:22SPEAKER_00: But then frontier models can do so much right now that there's a lot of things you can bolt on top for you know not like generating summaries or understanding things, or certainly in the process of building models using the frontier models to help go build you the right custom model is another that's how things work.
- 27:43SPEAKER_01: And then Greg, can you just run me through for I mean you've run me through it before, but for people listening, like what is the business model for you?
- 27:50SPEAKER_02: The core business that we have today is really selling our hardware and software uh to clinics in the US.
- 27:58SPEAKER_02: And so it's a subscription service.
- 28:01SPEAKER_02: We basically mail them out this briefcase, and in it comes you know, this portable device that they can put on their head.
- 28:08SPEAKER_02: It comes with a laptop and an amplifier, and basically uh our software is already embedded.
- 28:14SPEAKER_02: So we send it to the clinics and they pay us a monthly fee.
- 28:19SPEAKER_02: But the part that's really attractive to us is you know, not only is it these sort of software subscription margins, which are really healthy, but every time we do a scan, we get more data.
- 28:31SPEAKER_02: And so the database grows and grows.
- 28:34SPEAKER_02: And so, as I mentioned, like every month we're getting at least 1,500 scans and growing every month.
- 28:41SPEAKER_02: So then, you know, it's this nice sort of network effects where the more data we have, the more that we can learn, the better products and offerings we could provide to clinics, which then gets us into more clinics, which then gets us more data.
- 28:55SPEAKER_02: So it's this really nice flywheel network effect.
- 28:57SPEAKER_02: And basically we have these different tiers to our clinics.
- 29:01SPEAKER_02: So we mostly sell to neurology and psychiatry, the you know, the specialties.
- 29:05SPEAKER_02: Um, and over time, we really want to democratize access to this technology where we can eventually get it to you know primary care, et cetera.
- 29:13SPEAKER_02: But for now, we're sticking to the specialties of neurology and psychiatry.
- 29:18SPEAKER_01: Well, yeah.
- 29:18SPEAKER_01: And I mean, is it just is it just the scans that you get information on, or do you also kind of get to see how doctors are interacting with it and kind of improve the behavior of the AI that way?
- 29:27SPEAKER_01: Like you get kind of feedback by them being like, I mean, is it like Chat GPT or something where the doctor gets to talk to the AI about the image?
- 29:34SPEAKER_01: And so then you can kind of or what what is it, what is the experience like?
- 29:37SPEAKER_00: What we have today is really in a a report.
- 29:40SPEAKER_00: It's interesting when you're when you're in healthcare, because I think trying to figure out the right balance of what that should look like is is interesting today.
- 29:49SPEAKER_00: Healthcare moves slowly, and so you know, our model is today like a lab test.
- 29:55SPEAKER_00: So the most important thing is when you run a lab test and you get your blood panel, you get back your you know, you're getting back all these.
- 30:03SPEAKER_00: Yeah, that's basically the thing we're producing.
- 30:06SPEAKER_00: And then I think we're starting to experiment and and people are taking our report and putting it into Claude or ChatGPT and then you know interpreting on top of that.
- 30:18SPEAKER_00: We may build bake that into the product.
- 30:21SPEAKER_00: It you know, it's it's unclear today like what's the best way to do it.
- 30:24SPEAKER_00: But the key for us is what's that core report where you order this thing, you order the test, and you get back that panel of lab values that then you can do anything you want.
- 30:33SPEAKER_00: Because it's that core data that is really the asset people are going out.
- 30:38SPEAKER_01: And is is that that report is what gets generated?
- 30:41SPEAKER_01: So you've got by AI, right?
- 30:42SPEAKER_01: And that's your that's your model that generates that that data.
- 30:46SPEAKER_01: Okay.
- 30:47SPEAKER_01: Very cool.
- 30:48SPEAKER_02: Yeah, yeah.
- 30:48SPEAKER_02: So they do the scan in the clinic, and then it goes, uploads to the cloud, we crunch some numbers, and then send them this pretty report explaining you know all these different uh things in the brain.
- 30:59SPEAKER_01: And how how big is the market for this?
- 31:01SPEAKER_01: Like, do you know like how many clinic?
- 31:03SPEAKER_01: I mean, you have to know how many clinics are out there and kind of have a good idea of the TAM for just this this product.
- 31:09SPEAKER_02: Yeah, I mean, just for the specialties of neurology and psychiatry, uh, it's about 60,000 clinicians.
- 31:16SPEAKER_02: Um, and then once you get to primary care, I think there's 300,000.
- 31:21SPEAKER_01: Yeah, that's a lot.
- 31:22SPEAKER_02: Today, you know, we have a nice business in in a hundred clinics, and so we're we're just scratching the surface of this market.
- 31:29SPEAKER_02: Yeah.
- 31:30SPEAKER_01: It's so cool.
- 31:31SPEAKER_01: It's so cool to see how far you've come since we even first, I mean, we met like a little more than a year ago, Greg.
- 31:38SPEAKER_01: Um that was like shortly after you had just gotten, I think, gotten this done and got Dave on board.
- 31:44SPEAKER_01: Um, I remember these conversations and uh yeah, just really fascinating.
- 31:49SPEAKER_01: And then in between then and now, like the AI and the agentic space just took off.
- 31:54SPEAKER_01: And I'm curious, how do you maybe this is more for Dave, like how do you see agentic capabilities kind of reshaping the medical space uh going forward?
- 32:03SPEAKER_01: Because that's a whole nother step function, right?
- 32:05SPEAKER_01: Like what are you what are you seeing as how we need to design these solutions given agentic capabilities?
- 32:12SPEAKER_00: Yeah, it's it it's crazy.
- 32:15SPEAKER_00: Um, there's so much going on right now.
- 32:18SPEAKER_00: I think the biggest thing is that the in the speed at which we can experiment goes up.
- 32:24SPEAKER_00: It's interesting because some things don't move faster in the agentic age.
- 32:29SPEAKER_00: You know, an antidepressant, somebody goes on an antidepressant or they get a treatment, you have depression, like it still takes a long, it still takes a little bit of time to know if those treatments work.
- 32:39SPEAKER_00: Anything in healthcare, if you're trying to reduce somebody's heart attack risk, you can't do that in a day.
- 32:44SPEAKER_00: It takes a long time.
- 32:45SPEAKER_00: And so I think the biggest sort of nearer-term agentic area for me in healthcare, I think, is thinking about how do you be really smart about the experiments you do and the and the data you capture, and how do you use AI to optimally figure out your path for developing these so you squeeze all the information out of the data that you have today.
- 33:06SPEAKER_00: I think ultimately, like trying to reduce the number of mistakes basically.
- 33:10SPEAKER_00: Hey, can we figure out what are the right trials that we can do that'll give us the most data?
- 33:14SPEAKER_00: Because the worst thing that happens today is people will do a one-year or two-year trial and then it comes out inconclusive.
- 33:21SPEAKER_00: And that's like a nightmare.
- 33:22SPEAKER_00: And it can just set you back forever.
- 33:24SPEAKER_00: And the fact is, we can't reduce some of these trials.
- 33:27SPEAKER_00: Like sometimes it just takes a year to do a trial because you have to wait that long.
- 33:31SPEAKER_00: Like biology doesn't move at the space of AI.
- 33:34SPEAKER_00: And sometimes you have to wait for the biology to go.
- 33:37SPEAKER_00: But um But you know, that that answer also changes every every week for me because the whole space is changing so quickly.
- 33:46SPEAKER_00: So it's it's tricky to answer that.
- 33:49SPEAKER_01: Well, so we're we're getting to the top of the show here, and I always ask this what have I not asked you that I should have asked for either of you.
- 33:56SPEAKER_02: I need to think about that first.
- 33:58SPEAKER_02: That's good, that's a really good question.
- 34:00SPEAKER_01: Um it gets everybody.
- 34:04SPEAKER_00: Yeah, yeah.
- 34:05SPEAKER_00: There's a what's next question, um which is which is an interesting one for us.
- 34:11SPEAKER_00: We are mainly diagnostic today.
- 34:13SPEAKER_00: We are trying to do a better job of understanding what's going on in the brain, trying to do a better job of putting labels on the different things that are happening today.
- 34:24SPEAKER_00: But ultimately, people don't just want to know what's going on, they want to fix it.
- 34:27SPEAKER_00: And so the what's next or where does all this lead is ultimately we see, okay, how can we leverage all of this data and all of this knowledge to do a better job of guiding treatment and ultimately figuring out what are the right new treatment modalities to be uh to be backing, going after.
- 34:43SPEAKER_01: It's like whoop for brain health.
- 34:45SPEAKER_01: You know, eventually one day, like I'm gonna have one of your devices at my house and I'm gonna put it on like twice a week, and it's gonna and it's gonna tell me.
- 34:54SPEAKER_01: I'll have like a pattern of things.
- 34:56SPEAKER_01: I'm gonna, it's gonna be like Greg's voice inside the headphones.
- 35:00SPEAKER_01: Close your eyes now, you know.
- 35:02SPEAKER_00: You know, I I love the whoop.
- 35:05SPEAKER_00: There's like there's two things this has done across society.
- 35:09SPEAKER_00: Yeah, yeah.
- 35:10SPEAKER_00: Yeah, we all have it's like I need to drink less and I need to sleep less.
- 35:16SPEAKER_00: There's yeah, it took tens of millions of us wearing these things for years to finally figure out like, yeah, we really should be sleeping better and drinking less.
- 35:25SPEAKER_01: Uh yeah, my my my whoop age is at 33.4 and I'm 40, so that's pretty good.
- 35:31SPEAKER_01: Nice.
- 35:32SPEAKER_01: Nice.
- 35:34SPEAKER_01: For your brain, you know, for your brain, like that's the that's not accounted for at all.
- 35:39SPEAKER_01: There's no device that does that.
- 35:40SPEAKER_02: Yeah, there's nothing to do with that.
- 35:42SPEAKER_02: There's a couple like really interesting use cases that you know Fireflies had that that aren't necessarily in the clinical setting.
- 35:49SPEAKER_02: Like right now, we're in the specialties where you know someone has major depressive disorder or they're worried about dementia.
- 35:56SPEAKER_02: But we have some stories where uh someone scanned their brain and had anxiety, and Dr.
- 36:02SPEAKER_02: Hagerdon was like, Oh, you just need these magnesium supplements.
- 36:06SPEAKER_02: They took them, and then all of a sudden they started getting like an hour more of sleep every night.
- 36:12SPEAKER_02: And if you just think about how that like compounds and like your mental health, um, and their anxiety came down, they're getting better sleep.
- 36:19SPEAKER_02: And so, you know, we have tons of like really interesting, you know, stories and anecdotes.
- 36:25SPEAKER_02: And so, yeah, I'm excited for as we kind of democratize access to this technology, more people can use it and we can sort of optimize health, which is really fun.
- 36:36SPEAKER_01: Very cool.
- 36:37SPEAKER_01: Well, look, I really appreciate you guys taking the time to come on the show and would love to check in with you in another six months and have you on again, see how how Fireflies is doing.
- 36:44SPEAKER_01: And, you know, thank you for the impactful work that you're doing.
- 36:48SPEAKER_01: I think it's very cool how you guys got these technologies together inside one company.
- 36:53SPEAKER_01: Because that's the other thing that I think going back to my earlier question like how did you see this?
- 36:57SPEAKER_01: Like, this is not the normal story you hear, right?
- 37:00SPEAKER_01: Usually it's like I was a founder and I had this idea and I like fought and then I like got it.
- 37:04SPEAKER_01: You guys like found some parts and figure out a way to acquire them and like stitch them together to to make this thing work.
- 37:10SPEAKER_01: And I think that that is even more rare than succeeding at like building your own vision, like that was from the ground up, this thing you just pushed for for like a decade.
- 37:20SPEAKER_01: Yeah.
- 37:20SPEAKER_01: Um, it's just very interesting.
- 37:22SPEAKER_02: The the thing that I kind of explain to people is like to be in neuroscience, you have to be good at like at least 10 things.
- 37:30SPEAKER_02: And so, you know, when you think about it, it's like, well, you have to be good at at the science and be in you know the neuro background.
- 37:36SPEAKER_02: You need to be strong at hardware, which is which is a towns, software, AI, FDA, regulatory, reimbursement.
- 37:45SPEAKER_02: It's like, and that is your ticket, that's table stakes to play the game.
- 37:49SPEAKER_02: And then it's like finance, ops, HR, etc.
- 37:52SPEAKER_01: Running company.
- 37:53SPEAKER_02: Yeah, and so like when we were looking at these assets, we're like, hey, this one company's got really good data and algorithms in an FDA clearance and really good science, but they're missing all these pieces.
- 38:05SPEAKER_02: And then, like, hey, this group, they like they're really strong on the hardware and the science, but they sort of didn't really know how to run a business, you know, efficiently.
- 38:14SPEAKER_02: And so when we started looking at them, we're like, hey, we can start putting this together, you know, the sort of the missing puzzle pieces.
- 38:20SPEAKER_02: And we think there's a whole bunch of interesting puzzle pieces that we can start to add to this uh that we think can sort of accelerate that.
- 38:26SPEAKER_02: So we're we're really excited about the future.
- 38:29SPEAKER_00: It was tricky at the beginning when we're like all of these companies have tried some variant of this and they've all failed.
- 38:37SPEAKER_00: And so the easy conclusion out of that is this is a bad idea and it won't work.
- 38:43SPEAKER_00: And it was only when we actually went through all of those companies and started analyzing why they failed to realize it was all different reasons.
- 38:52SPEAKER_00: And if you put them all, like every problem had been solved somewhere by somebody.
- 38:58SPEAKER_00: It's just they hadn't all been solved by one company in one place.
- 39:02SPEAKER_00: And so there was a pretty big leaf of faith at one point to be like, oh no, we're just gonna succeed.
- 39:09SPEAKER_00: We're gonna be able to succeed where all these other people have failed, because you know, there is no bad idea here.
- 39:15SPEAKER_00: It was just nobody ever put it together right.
- 39:17SPEAKER_01: Sounds a lot like that conviction and leap of faith to bet on that competition and that you guys would win.
- 39:22SPEAKER_01: I think there's a pattern there, you know.
- 39:25SPEAKER_00: Yeah, that one I think was easier in a way.
- 39:27SPEAKER_00: That was simply betting on, you know, that was that was maybe just a little bit of hubris and being young and not like, ah, we're just we're just the best at this and we're gonna win.
- 39:36SPEAKER_00: Um this one was a little scarier.
- 39:39SPEAKER_01: Awesome.
- 39:40SPEAKER_01: Well, I really appreciate you guys coming on the show.
- 39:42SPEAKER_01: Thanks so much.
- 39:42SPEAKER_01: All right, perfect.
- 39:44SPEAKER_01: Thanks for having us.
- 39:45SPEAKER_03: Using that AI analytics layer to actually scan against all sorts of options and rule out the ones that don't make sense is just to me like chef's kiss.
- 39:56SPEAKER_03: Great job, guys.
- 39:56SPEAKER_01: Yeah, and I think I this is a really great example.
- 39:59SPEAKER_01: And we've had a couple of them on the show.
- 40:01SPEAKER_01: We recently had Red Sky Health, which that one, I think that episode came out recently.
- 40:04SPEAKER_01: Both the things these two companies have in common, they're both in the medical space and they both got access to data that you wouldn't otherwise get access to.
- 40:12SPEAKER_01: Like, this is actually very hard for other, even some of the big labs to get access to this data.
- 40:16SPEAKER_01: For entrepreneurs who might watch our show, you know, you should watch out for these patterns if you're looking for a place to build.
- 40:22SPEAKER_01: If you find data silos or data that's like kind of difficult to aggregate or access, that is a huge opportunity.
- 40:28SPEAKER_01: I think everybody's kind of chasing that like holy grail of a general purpose agent that like runs your life.
- 40:33SPEAKER_01: And I think there are like definitely hundreds of thousands of opportunities like this where there's specific data that if you you you can fine tune a model on and you can use open source models probably to do it and really be effective.