#208 Michael Martin: How AI is Saving Lives Every Day (Inside RapidSOS’s Groundbreaking Tech)

0:00There are plenty of examples where Rapid SOS is driving a 5-7 minute response time, and where a traditional alarm system is driving. I mean, the data from a couple of weeks ago is an example, it's 52 minute response time, right? So you can start to really start to drive significantly faster, more effective response. You know, if you're using some sort of device, security system, et cetera, that has RapidSOS inside of it, if you almost like Intel inside. In this episode, I have the pleasure of speaking with the CEO of the intelligent safety platform, Rapid SOS. Michael Martin. Rapid SOS is an intelligent safety company that harnesses artificial and human intelligence to fuse life-saving data from 540 million plus connected devices, apps, and centers from 200 plus global technology companies to over 21 ,000 public safety agencies in six countries.

1:05The company works with hundreds of top companies, including Apple, Google, Uber, Instacart, and Honeywell. I hope you find the conversation as fascinating as I did. Well, great to be here, Craig, and thanks for having me. So I'm Michael Martin, one of the founders and the CEO of RapidSOS. And RapidSOS is what? So RapidSOS is an intelligent safety and emergency response platform. So what we do is over half a million times a day, we fuse billions of sensor feeds into one unified picture of response for 911 and first responders. And so that manages emergency response for about 540 million connected devices or over 99 % of the U.S.

1:56population. Wow. Yeah. And we were talking earlier about how large this problem is and no one's solving it in the age of AI. Go back and tell me how you got involved, what your background is, and the scope of the market, so to speak, that you're addressing. Yeah, so I grew up in a rural farming community in Indiana and had hardly left the state. And then it got that first big job after college, came to New York City and was, you know, like a lot of young people in New York City, was working very late. So it was like two in the morning, got off the subway in Spanish Harlem where I was living and had my welcome to New York mugging experience.

2:48Craig, first time in my life I'd ever thought about calling 911. And the first time I realized in the middle of an emergency, like how hard it is to like, get out your phone, dial a number. And like, you know, like this dude's like kicking my butt. It's like, time out. I just got to make a call real quick. And oh, how many ambulances are we going to need when you're done kicking my butt? 911 needs to know, right? I mean, it's just crazy to think about like how this system works. And the other thing, which I had no appreciation for Craig was the scope and scale of that challenge. Last year, there were 250 million 911 calls.

3:21That means statistically, three out of four Americans called 911 last year. Now, look, it's not evenly distributed. There's pre-existing health conditions, domestic violence, a variety of reasons that cause some folks to access emergency services more than others. But still, this is this ubiquitous service we all rely upon. And in the worst moments of our lives, with all the technology around us, smartphones, phones and AI and video and all this sort of stuff. 9-1-1 often doesn't even know your name or your location. And the consequences are profound. Last year, one out of three American fatalities occurred on the phone with 9-1-1 or in the moments afterwards.

3:58One out of three American fatalities. And it's, you know, if you put that in perspective, a million out of 250 million, it is remarkable how well the system does work. So, you know, despite the volume of emergencies, but that is a testament to the people of 9-1-1, not the infrastructure, not the technology, Craig. So literally, this was a problem that, you know, smacked me in the face in 2012 in Spanish Harlem and ended up impacting actually my father in our, you know, where I grew up in rural Indiana a few months later. So I went to grad school. My co-founder was doing his PhD at MIT. I was in grad school at Harvard.

4:34And so we spun out of grad school in 2015. And since then, the business has raised about$350 million that we've invested into building this unified AI platform that today is supporting most emergency response across six countries. And I'm just curious, is this something that is applicable to other markets outside the U.S.? Yeah, you know, last year we supported 171 million emergencies. Today, as of this year, we're now tipping about 550 ,000 a day. But that is still touching less than 10 % of the total volume of emergencies globally, Greg. Greg, so this is unfortunately a reality in the world that we live in, right?

5:22With aging populations, global pandemic still in some cases, you know, war and uncertainty in other parts of the world, climate change, rising violence. Like there's just all these challenges that, you know, I think it just kind of breaks your heart that in the day with all the ways that technology has transformed our lives, like how is it, right, that we aren't having a greater dent on emergencies? And that is really bringing down, you know, longevity numbers and, you know, like, like, you know, people my age still the leading cause of death is car accidents. Like all the safety tech, you know, cars now have automatic braking, lane keep assist, all this sort of stuff.

6:01And yet the fatality rate in automobiles has largely been flat or actually was even higher post, you know, post coming out of the pandemic. It's, you know, out of hospital cardiac arrest, 359 ,000 fatalities a year in the United States, 1 ,000 people a day die from heart attacks in the United States. So there's just like in a day and age where wearable devices can detect heart attacks now, right, where all the sensors exist. It's just like, how are we allowing that to occur? And so I think we're on the cusp of just this extraordinary moment with AI and all these connected devices to truly transform safety, security, and emergency response.

6:39Yeah, that's interesting. Again, on a personal note, I have a brother that fell recently, and he now, you know, I'm getting to an age where falls are a big deal. And he now wears an Apple Watch that I think is programmed. It'll ask him, is this an emergency if he says yes, and it'll call 911. and 911. So tell me how you knit together all these devices and what devices are they? These companies are working in harmony with 911 and first responders through RapidSOS to facilitate the fastest, most effective response possible during an emergency. And so we can think about this as previously that was effectively, Craig, an analog copper phone call in.

7:34And so you You needed to call and then you needed to verbally articulate your name, your address. Often you would need to even spell your address and then explain exactly what was occurring. Right. And so if we digest that, there's there's kind of three foundational challenges with that. One is human detection. Like as you pointed out, often sensors actually are now detecting incidents before humans do. So, you know, if there was a fire in this building right now, often the fire alarm is going detect that before me as a human might detect that right or you mentioned kind of an apple watch detecting a fall or a car detecting a crash right so one is rapid detection and let's just remind ourselves you know in the example of of the leading cause of death in the united states out of hospital cardiac arrest every minute in response time craig is a seven to ten percent reduction in mortality rate so seconds and minutes matter so first we got to detect stuff immediately and we We need to use our sensors to do that.

8:33Second challenge is how do we integrate that into these really kind of complicated operational environment of 911 and first responders? And if you don't do that well, you can actually yield really difficult and challenging results. So an example of that is many of us have home security systems in our home. So actually, one of the largest single blocks of 911 traffic in the United States is home alarms. But 95 % of those alarms are false. So put yourself in the shoes of a first responder now. So, you know, you could work five years and only respond to false alarms. And then one time you show up and it turns out, you know, it is actually a real incident.

9:14There's a bad guy in the home with a gun. And, you know, recently I was looking at the body cam footage coming out of one of these examples where fire department comes rolling into a homeowner's house. Police department shows up to. Again, these folks are on different systems. So first they're talking like, what are you doing here? And the, you know, the fire guy's like, we got a medical call. One of those pendants, right? The fall help. I probably can't get up pendants. The police officer's like, I got a bird alarm. He's like, man, this guy must've hit every single button. Let me go first. Make sure it's okay.

9:42Officer goes up, rings the doorbell. No answer. Shines his light through the door. Doesn't see anything. Walks around to the backyard. Shines his light in. This was in Texas. Okay. Homeowner at this point in time, finally starts to wake up, sees a guy in his backyard shining a flashlight in, grabs his gun. Okay. Officer all of a sudden sees in the home a guy running out with a gun. Officer draws his weapon, fires, thankfully misses, runs out on the radio. Shots fired, shots fired, backup needed, all this sort of stuff, right? Police come streaming in from all directions. And here like five minutes later, out walks a 72-year-old homeowner in his boxers.

10:22What are you all doing here? I could have killed you. Like, you know, and like, this is the challenge, right? So like, even when we detect incidents, right, we've got to operationalize the data, verify that information is real, unify it into the systems that first responders are using every single day. So that's point number two. Then point number three is I really want to challenge us actually to think expansively here, meaning that like, like, do we really have to always be reactionary to emergencies? Is it really such that like emergencies just happen and then we've got to figure it out? Or does snowfall cause car accidents?

11:03Does speeding and texting cause car accidents? Right? You know, are there things we can learn for that? So when we were in grad students at Harvard, we actually studied how weather causes heart attacks. And so it turns out like, you know, many emergencies, we believe are actually going to be predictable and preventable. And so you can imagine a future, right, where that wearable device actually tells you, Hey, Craig, like, sit down, you know, you're going to have a heart attack in the next five minutes. It's already good. You know, we've already got the ambulance on the way. Press here if you want to just go ahead and notify your loved ones, You know, everything's going to be fine.

11:39You know, that that sort of instate. So so step one, detect that stuff automatically, immediately. Step two, fuse that sensor seeds, all that information directly into the hands of 911 and first responders. And then step three is learn from every single one of those incidents so we can start to predict and preempt these things in advance. Yeah. The sensors that you're talking about, is is does it go beyond a wearable? Yeah, so we power over half a billion connected devices. And so that's going to look something like 10 million connected vehicles. We're powering fleets of trucks now, something like 10 million camera feeds, obviously wearable devices, smartphones, home security systems, commercial building alarms.

12:26So there's a growing, in fact, I was just talking to a baby monitor company as an example, right? So I have a four-week-old at home. So it's personal to me that like, you know, SIDS obviously is a terrifying thing for new parents and the way, you know, now there's tech that can detect if your child stops breathing in the middle of the night. But then the next challenge is what do we do from there? Right. And so, you know, if you can instantly get that information into the hands of your local paramedics, if you can use the camera feed to coach the young parent like myself on the CPR. And we can even imagine ultimately an in-state where as that paramedic, the front porch lights go on as they show up at the front door, they tap their smartphone, the door unlocks.

13:07You can see how this ecosystem really starts to build to totally transform outcomes. Yeah. I have to say I have a couple of directions I want to go. But one, you talked about how 911 calls are not spread evenly across the population. wearable devices are certainly not spread evenly across the population how do you get sensors into the homes of people who who can't afford an apple watch i mean that's not your problem but how do you guys think about that going going looking into the future yeah you know i I think we've longed because of our roots coming back to. So I grew up in one of the poorer parts of of of Indiana and a rural farming community.

14:04And to this day, right, folks were in my hometown. Right. Generally would not have the same sort of technology that, you know, for example, like my neighborhood in New York City, where I'm talking from today would have. Right. So so like figuring out, like, how do we upgrade the foundational infrastructure so that every single connected community? device can harness this technology and capability set was pretty critical to kind of that work. So the first thing we did, right, is we provide our core service and platform out to every single 911 agency for free in the United States. So today, as I mentioned, we power 21 ,000 agencies.

14:39We do have some paid products on top of that if they want to do more with that. But the baseline service is for free and now it covers 99.9 % of the US population. So almost certainly if you are living in the United States, like your local public safety agency now has, you know, all this richer sort of capabilities. Then the next thing, right, is we work very closely with connected device companies so that this is foundationally embedded inside their technology. So it's not, you know, it's often not some sort of additional service you have to subscribe to or pay for, et cetera. So like, you know, if you have a smartphone, this is just embedded out of the box with that device.

15:17Now, look, we rely on modern connected devices. So if you are on, you know, like a legacy sort of phone, you know, there are going to be limitations, unfortunately, today. But, you know, another example of how everyone can help first responders for free is we partnered with the Red Cross and the American Heart Association and Direct Relief to standup emergencyprofile.org. And so this is a free service that anyone can go and create a health profile on. And these things, as we talked about, Craig, are really powerful, right? Like letting first responders know what are your preexisting health conditions?

15:53What are your medications? And, you know, and if you have some sort of disability, like, I mean, once again, you just put yourself in the shoes of those police officers. You've been called that, hey, there was some sort of gun violence or something like that. And you show up on scene and you're providing instructions to someone you see and they're unresponsive. They're ignoring you. Right. What does that mean? Well, in a normal situation, police officers train like this is a threat. But if they knew that you were deaf or hard of hearing, it's totally changes how they, you know, how they operationalize that situation.

16:26Right. Or if your kid is autistic or things like that. So, yes, this is something like we want to make sure that every single person in the United States and ultimately globally can access this technology and service for free. It is available to all of our first responders for free. And it's, you know, inside our connected devices today for most of us. Yeah.

16:49And let's give listeners the name of the company and the name of the products. I know there's RapidSOS and Harmony and sort of give me the mental map there. Yeah, yeah, yeah. So RapidSOS is the name of the company. And the way we talk about the platform is an intelligence safety and response platform, right? And so what we mean by that is if you think of it like visually, like a horizontal platform, right? So on the left side, we're funneling all this incident information that is being detected by connected devices on the edge. And one way you can know if your device is enabled or not is you can look for a little rapid SOS ready badge on on the device.

17:40Or obviously you can ask the device manufacturer if that is a capability. And this this turns out to matter pretty significantly in our own purchase decisions, whether that's the consumer or the enterprise. So I was just looking at response data for New York City, where I'm based and talking to you from today. And look, there's variability and complexity in this data, Craig. But if you look at it, there are plenty of examples where Rapid SOS is driving a five to seven minute response time and where a traditional alarm system is driving. I mean, the data from a couple of weeks ago is an example.

18:15It's 52 minute response time, right? So you can start to really start to drive significantly faster, more effective response. You know, if you're using some sort of device, security system, et cetera, that has rapid SOS inside of it, if you will, almost like Intel inside, driving that more effective response. So that's on the input side of the platform. On the output side is the technologies that public safety agencies interface with. So we are embedded in over 4 ,600 public safety software systems. So the great thing for public safety agencies is, you know, we're already inside the systems that they're using.

18:53We do have a flagship product that about 90 percent of 911 agencies in the United States are using today. It's called Rapid SOS Unite. And it is a unified picture of all your emergencies going on in one data rich map with all this contextual information effectively. And then we also have a responder app, which is used by over half a million first responders in the field, which is called I am responding. So that's, again, left side on that platform, all that data coming in. Right side, Unite, I am responding is how that data is being operationalized. Also inside every major public safety software vendor.

19:30And then you mentioned Harmony. So Harmony is an underlay AI layer which powers that whole platform. And that's, you know, we've been working on Harmony for a long time now, really, you know, since we were spinning out of grad school. And, you know, last year we tipped over half a billion emergencies supported through this technology infrastructure. And I think we're really seeing this powerful moment for how AI can really serve as that assistant to 911 and first responders in these moments when seconds really matter. So Harmony is a combination of foundational models as well as integration of some of the new generative work to effectively structure all that data, those billions of data payloads coming in on the left side, structure that into 911 and first responder systems and really serve as almost like a co-pilot to that 911 telecommunicator.

20:30communicator yeah um i i the other thing i wanted to ask about um i i had a call with a researcher named feifei lee at stanford a couple of years ago and she was working on something called ambient intelligence and which which in her uh the you know research was connecting sensors to a central ai model that could learn a behavior she was thinking of it in terms of elder care that could learn the behaviors of of the people in the if it was in the home or in the in the hospital room and would detect any anomalies you know if if it's if it's an elder elderly person living alone and they go into the kitchen every morning and make coffee If they don't, it detects that and then would send a note to a loved one or a health care provider.

21:50Could this be adapted? Have you guys thought about that, about taking it beyond the emergency services world? world. Yeah, I am really excited about all the work that is going on right now to think about enhancing that detection of incidents. We work with a number of companies in that space, companies that do things like weapon detection, gunshot detection, learn about behaviors in the home and where someone might be facing an emergency, right? And we do have a suite of solutions now that manage situations where we aren't yet sure that we need 911 first responders, the ambulance or the SWAT team.

22:32And we need to just confirm, is this a safety incident? Are there other types of care teams to loop in on that particular situation? So we do power some solutions in the remote patient monitoring space, as an example, Craig, that are similar to what you're describing, where there is sensors embedded in, say, a hospital at home environment or a medical discharge situation or an aging in place environment where the sensors are creating that map and picture of human behavior in the home. And if something is anomalous, it's a way to check in on the individual, make sure that they're supported and cared for.

23:04And obviously, in any situation, immediately roll an ambulance with all that critical data if it's warranted. Yeah. So let's talk about the science. science you mentioned uh foundational models of course everyone's building on foundational models is are you accessing uh models through api uh and who are you using uh and then above that you fine-tune the model for your purposes and the data coming in can you just talk about that infrastructure. Did you ever think why some businesses are all over the place and you're familiar with their brands and you interact with them seamlessly? When you think about those businesses whose sales are rocketing, like Feastables or Mr.

24:01Beast or Thrive Cosmetics, you think about an innovative product, a progressive brand and button-down marketing, but an overlooked secret is actually the business behind the business, making selling, and for shoppers buying, simple. For millions of businesses, that business is Shopify. Nobody does selling better than Shopify. And the not-so-secret secret with ShopPay that boosts conversions up to 50%, meaning way less virtual shopping carts going abandoned and way more sales going. So if you're growing your business, your commerce platform better be ready to sell whatever your customers are scrolling or strolling on the web, in your store, in their feed, and everywhere in between.

24:58Businesses that sell more sell on Shopify. Upgrade your business and get the same checkout that Feastables uses. Sign up for your dollar per month trial at shopify.com slash Ionai. That's all lowercase. Shopify, S-H-O-P-I-F-Y.com slash IonAI, E-Y-E-O-N-A-I, all run together. Go to Shopify.com slash IonAI to upgrade your selling today. Yeah. So, you know, I think we started with a more kind of traditional machine learning sort of approach. And so we were not on the bleeding edge of all the work on large language models. But what has been really exciting is obviously the availability of that technology today.

25:56So we already had a foundation where we were looking at what were all the questions that 9-1-1 needed to ask to figure out a response and how could we learn from every single one of those, right? So if if you if you study an incident, you look at 911, you know, that human intelligence said, hey, we need to dispatch these units with this priority level with these pre-arrival instructions to this location. Right. And what was the chain of questions and decisions that got us there? Right. And is there a way for us to surface and pull information out of these sensors so that we can serve effectively as that that co-pilot for for that 911 telecommunicator?

26:35Now, the Harmony platform, or, you know, once it is with the proliferation of large language models, we were able to really, I think, accelerate the ability to operationalize that information into the hands of 911. But then we found in testing, not surprisingly, Craig, that we were facing real challenges with hallucinations. And so we had to think about obviously training the model. We had to think about the right temperature of the model, really having a low temperature sort of approach. Right. So trying to minimize any sort of creativity in the model. And then, you know, I think the biggest innovation, which is now becoming more mainstay for everyone, is really a retrieval augmented generation approach approach to this.

27:19Right. So so I'll just give you like an example of how that works. So we one of the new challenges facing first responders is all the electric vehicles that are out there today. So you have it turns out it takes massively more water to put out an EV fire. But more than that, other tools like the jaws of life. Right. Like when you have, you know, the new Tesla Cybertruck and 800 volt electric line running through there. Right. You just can't just cut into this thing like you could in a traditional internal combustion engine. Right. And so, you know, if you go to these various automotive only manufacturers websites, there's just like hundreds of thousands of pages.

28:01I'm not kidding, Craig, like between them, hundreds of thousands of pages of documentation. So the way it's supposed to work is you get into a crash with an EV. Today, someone calls 911. They say, hey, I was in this crash. First responders show up and they're like, oh, I think this is an EV. Somehow they read it's the new Chevy Silverado EV. They go to General Motors website where there's, you know, it's not even like every Every model year has a different manual on how to handle it. Every version often is different. So we got to figure all that out, figure out the right one, flip through 150 pages of documentation, figure out exactly how we're supposed to do all this while someone's trapped inside in a fire.

28:41Right? I mean, it's just crazy. So, you know, if you take a traditional, you know, large language model approach to this, that's been trained on a more general use case, right? you start to get really dangerous hallucinations because you get situations where we're taking, you know, Reddit threads and things like that. Right. And we're using that, you know, and the challenge, right, even in that stuff, is there's real nuance between again, is it, is it, you know, like, like even in Tesla, right, the battery chemistry in a model Y will vary based off a model. Like it's, it, it, there's real specificity on this.