Health Tech Entrepreneur: Vincent Hartman '08, MS '22, is founder of a startup, that makes AI summaries of doctor notes, for better healthcare outcomes

[0:00]Hello, the guest today is Vincent Hartman, class of 2008, Masters Science Class of 2022. He is the founder of a health tech startup in New York City. So hi Vincent. Well, what does your startup do? Nice to meet you, Tony. I'm Vince Hartman, Co founder and CEO of Abstractive Health. We help clinicians know their patients before the visit, so we have the capability of retrieving a patient's full longitudinal medical record. We then build a complete understanding of that record via AAI medical record summary and then we give key insights on that patient so that the doctor walks into the exam room knowing who that patient is. Interesting. So, so from the patient perspective, what does it look like? From the doctor perspective, what does it look like? So from a patient perspective, very often in the United States, you will like complete forms for your appointment and you walk into that your doctor and they don't know much about you because they just don't have the time to study up on your record. Your record can be hundreds of pages and physicians have maybe 10 to 15 minutes of prep and they take a lot of information just of what you provide them. So if you like, have a prior medication from years ago, that might be impactful for that visit. You might not know yourself that you need to bring that up or that like maybe an allergy to penicillin is impactful. With abstractive, we give the clinician the ability through your consent, through the patient's consent to understand and to retrieve that information. So that when the patients and the doctor have their that visit, that the clinician is up to speed on that care and that they can know you better. And so that they can make sure that that medication we were just mentioning is in their mind that they don't do a

[2:02]contraindicated like result for it. Gotcha, gotcha. That makes sense. So so I'm trying to think of like there's input to the system and there's output of the system. Like what's the input into your company and what's the output? Of your company. So we are a direct implementer of the National Health Information Exchange System in the United States. So there's a layer above the computer systems that hospitals and your doctors use. Those are often called electronic health records. Common ones are Epic, Oracle, Athena Health, there's about 300 like EHR systems and over the last 15 years the country created a National Health information exchange so that it facilitates the record sharing so that the doctor is able to pull all that content. There's about 20 to 30 companies that facilitate the record sharing process on carry quality. And then there's another one called E Health. And then the government has been pushing for an exchange system called TEFKA. So we have the ability to retrieve a patient's longitudinal information on this exchange. There isn't a national database of patient records. It's like stored in all of the doctors like individual practices we query out, we get all these records and then our system will parse and use large language models to create an AI summary that synthesizes in about like 800 words from the millions of tokens that are like in the record. And then the clinician can ask questions, get like answers on that record. And so they're able to review it quickly in a software that we have in like the browser for the clinician. Like, like for a patient like who has like 100 pages in their document, like what are the 100

[4:04]pages like? Like is it like pictures of their heart? Is it like food that they ate at the hospital? What? What's in those pages? So there's definitely images of X-rays, heart images, DICOM images. Often like when you go to a clinic visit or the hospital, the clinicians are after they care for you. They write a note of the like what they saw in your visit and like objectively what they measured. And then they like do the assessments like vitals, labs are listed. And then that information becomes accessible so that they can track over time how like different physicians are documenting on your care so that when the consults, like maybe you go see that cardiologist that you had prior saw, like your primary care physician and all the other clinicians, but they're not starting from scratch. Like all those records are accessible. And then patients have the ability nowadays to retrieve that content themselves, to review it and to make sure that it properly represents like the care that they expect. Gotcha, gotcha. And so there's like for as you get older, you will have hundreds to thousands of these documented notes that are expressing like, like when you go to the hospital to have a baby, like it list out your delivery. And then like when you get your colonoscopy, it list out that as you like get into your late 40s. And so these like different, all those different outcomes are getting measured. Gotcha. So it sounds like this health exchange is like a really good solution. Then it sounds like from all the other health tech entrepreneurs I talked to, it seems like the health tech data is like the hardest part to get. Well, like what is hard about

[6:04]the health tech data that you face? So there is accessibility for clinicians to have a first right access to get data for patients that they're treating. And then patients, now there's a push that patients also have a first right access to get their own data. In the healthcare ecosystem, there's a lot of different players that are not doctors and that are not patients that would value getting patient data. Those different players do not have a first right to get the data. So like a like a company that's focused on like clinical trials, they cannot access from the exchange your patient data. There is secondary uses of data that like if the patient consents or if like the hospital like they can use like, so like hospitals will have clinical trial components, but they had the they first did treatment on you and it's like, and so the challenge with a lot of digital health companies is that they need data, but it's not necessarily initially focused on that treatment workflow that the exchanges help facilitate. So they don't have first right exactly have the first right and that's like the so you are the first right layer of. The yeah. So then our product, we predominantly we only are working with clinicians because that is who has like that. We verify that the user on our product is a clinician with an NPI that they treat patients like I said. And so then when they sign up on to obstructive health, we know that they are using the data for treatments because we've confirmed we do, we work with a company called Vouched. It's similar to like ID dot me

[8:08]or Clear. It's in the identity verification space. And so we do identity verification of all of our users that sign up. Got you, got you. So in the first, so for the doctor as well as the patients want to make sure that you're getting the right data from the right person. Yeah. And so then the clinicians during the patient visit get HIPAA consent and then HIV retrieval consent based on state law for pulling the patient record. And so then via that consent, they are able to capture the content and obstructive health. So how do you think of modes in your business? Like there's like the so, so you have like the access to the exchange that's like hard to get. That's like a mode. Like what other modes are there around the AI model, around the data? Like what modes are there? So there's as you mentioned, the data access mode and so there's. It takes about two to three years to get regulatory access on the national exchange. And then there's just an ability to do that well. So medical data is very messy and so you have to parse it. We use OCR to take the images, to grab the text off of it. And we also we have a Moat of how we do our summarization. So we have patents and we have 4 research papers at this point on our medical record summarization and what we spun out from Wild Cornell Medicine. So we were, we did research for two years and then we've been at it now for four in proving our like efficacy on how accurate we are for creating a medical record summary that it's safe for the clinician and then it properly synthesizes the information in a reliable manner. Well, like like what layer of like the AI inference really were you working on to create this like these patents or are they on like the rich layers which? So we do a summarization

[10:10]approach that is called like section based summarization where we take the collection of documents and we do mini summaries on every individual documents and then we link everything in that to the original source documents. We collate all of the mini summaries and then we create a larger meta summary and then that meta summary is linked to the original. So that like when literally every sentence that we have in our final summary, it links to the sentence of where it came from in the original document. So we have we have very like deep source verification that we go a step further. So like a lot of AI companies will do reference or citation based sources where they'll like act similar to like a journal article. They'll say a sentence and for a few of them, they'll have like numbered citations that link. And so we go a step further that literally every sentence is embedded with the sentence it came from in the source documents. And that logic, that UI component, we have patents and intellectual property on that process. Gotcha. How does it deal with like? If like there's a contradiction within the source documents, How does? It deal with contradictions. So we will like we have some flagging that we inform users of contradictions, but more we focus on the original presentation and salience of the source documents that we don't attempt to infer on what is the most appropriate answer to the contradiction. Because then that would in

[12:12]itself be doing medical diagnostics that if you have like 1 instance that the patient has documented like this diagnosis and then you have another one that probably contradicts the original diagnosis, we present in a clinical event format and in the summary the contradictions. And then we, it's, we have like trained clinicians that are using the product and they can as they read, they like see the history of like the patients like story and so that they value the actual, the contradictions themselves. And so if they see how the patient is probably struggled in their story of clinical care and like how some clinicians got it wrong. And so we've actually found that they appreciate just source linking and not attempting to like with some of you cleaning up the record because then that runs into issues that. Makes sense. And then like over time, as more people use it, like does it like over the years, does it like get better over time? Like how does it? Improve. So we are consistently improving the summary itself and as users bring up and any sort of software, as they bring up bugs, as they bring up support issues, we make improvements of it. We take that knowledge and we don't use the like the data from our customers to train and fine tune the models. We built the original from a research partnership that we had with Walt Cornell Medicine. Since then, we haven't, we don't use the data for model training, but we learn from our customers on how to like improve the product from AI level. Gotcha. Very cool.

[14:12]And then and then so, so, so like 1 main worry in like AI companies that the model layer won't start, start encroaching upwards. Like is it because you have the health exchange rate that you're able to defend against the models coming up? So we, our whole stack is a state based structure. And so meaning that for our large language models and our AI that we have it do inference based actions. So like for one that we're talking about is summarization. Another common in AI is retrieval augmented generation or RAG that you use to like ask a question, retrieve relevant content, and then do language model to produce the best answer. That level of AI that we do has a like we need this task done and it's like we know it's doing that task and then we it will infer the best answer based on that. We do not yet have like allowing an AI would come, people would refer to as an agent that it sits and chooses itself to, for example, like call the patient. Like let's say you retrieve the records and you see on the patient's record a follow up for like a patient visit and they hadn't like scheduled this patient visit yet. There's companies you could consider doing an agentic approach where you call the patient automatically. We in our system have that we would like have more of an if then that like the the user would need to Click to inform the call the patient and not

[16:16]choose to call the patient itself. And so like what most usually say like AI moving up the stack or doing things that's uncontrolled. It's like the the clinical user is like interacting with the system and then they're asking like, why did it call the patient? And then you're like looking and you're like, well, the agent said it was best to call the patient. And then you're like, well, why? And you're like, well, if I look through it, it's it, it inferred. And there's no like if, then it's like the LLM chose. It was just best to do it. Was deterministic is good in this. Case usually yeah, yeah, it's very like it becomes harder for clinical users to logically think of why did this thing do what it did and then you as an engineer are like, well, the AI thought it was mess. That's that's the part that I think a lot of people in healthcare are not like. So usually most even in healthcare when people are thinking of agentic workflows, it's usually has still state based and there's like there's not much like true true agentic AI that is occurring without an if then logic that I can say why this happens. It hasn't like in healthcare yet moved up uncontrollably. I haven't seen it, and we're not doing that either. Yeah, and people want the doctor to call them. They want. Exactly, Yeah. Human touch. That makes sense. So there's a patient and there's a doctor. What about like the like the owner of the hospital or the chief operating officer of the hospital? What are their perspective on this software? So most of the like owners were the like you have like a chief medical officer, you have a clinical informatics officer, the chief operating officer. Most of these individuals themselves that work at large health systems or doctors themselves and they understand they deeply value the concept of one of the challenges in general

[18:20]that electronic health records have not distinctly solved over the last two decades is this problem that clinicians just don't know their patients and they, the doctors don't have time to find like to look through the data. And for that like risk problem that it's better to use AI to get a better understanding of a patient, it is safer. And there's studies that show that if you use a synth like you use an AI to synthesize. So you can quickly understand that that produces a higher understanding of the chart then if you did not do so. So you can like do a study where you have a doctor using EHR, no summary and then you have a doctor use a summary and then afterwards you ask him like, let's say 5 questions on the patients, like they, they're, you will get the doctor to know more about that patients in that period of time. And so the hospitals are very like excited and interested in this because they're measured on patients having good outcomes when they leave the hospital. So they don't want them being readmitted. Common health problems are like congested heart failure, sepsis, AK acute kidney ailment like so it's by knowing problems of the patient, you can solve them better you can treat the patients better and then they don't readmit you have better just general outcomes. Everyone in Healthcare is like all of it is like interwoven in this concept that in general by. Treating the patient better and knowing the patients, then it

[20:22]leads to a more like more revenue better like it helps everybody. It's like, it's like a just a fact of the healthcare ecosystem. Well, like from the owner of the hospital's perspective, like what's the algebra equation that they have to go through when they're considering whether to license the software or not? Is it like, so there's a cost of the software that's like the on the minus side, like what's on the positive side to balance it? So some things that like large health systems are considering from like revenue loss or revenue like potential that they get. So that was mentioning if you like the hospitals are reimbursed based on a generally a flat fee on the disease that you have when you are treated. And so if you, for example, go to the hospital and you have pneumonia, there is a code that they will complete to say we treated this patient for pneumonia. And they will get reimbursed for like, let's say $15,000 for that treatment of pneumonia. And if they keep you in the hospital for three days or if they keep you in the hospital for 10 days, they still get $15,000. It's a flat feed like you as a patient, your insurance company will bill you separately than how the Medicare or the insurance is reimbursing the hospital. Hospitals are reimbursed a flat fee and the structure generally and the structure is to incentivize to keep the patient in the hospital less time because you the insurance in the government, there's a lot of data that you get, you can get

[22:25]sick if you're in the hospital longer and if you if you can do treatment in three days as opposed to 10 days, it's better for the health like ecosystem for the patient for reducing cost. And so by using a system like abstractive, you can demonstrate that like by knowing that patients have these like during the stay and the clinicians treating their patients better, that they're able to not keep them as long and so that they reduce like the time at length, what's called the length of stay. So that's one measure. More information helps you reduce the length. Exactly. Another measure. And so like I caught like a very common problem in that hospitals deal with this called congested heart failure. And if you have congested heart failure, you will be above what's called your dry weight. And so like let's say you're typically like 170 lbs. If you have congested heart failure, you could like take on fluids and you would go up to like one maybe 80, not five, 190 lbs in that your wet weight. Yeah, yeah. So that that £20 if you know that you can order medication to like a diuretic, like to get rid of the fluids, which is better on the heart and you can released the patient from the hospital faster. But you need to know that the patient is not inherently overweight or at least that they have a baseline dry weight and that you they're above that from a fluid retention problem. And so knowing that faster like

[24:27]it prevents them from lots of other illnesses. And so if you like have that information, you can treat patients better from congested heart failure and all that is just like you know the patients and their information and treatments and so and then you they don't come back to the hospital. And so it's like all related to care. And then there's lots of other these measures you can give me going. So for like the obstructive health like for the congested health thing like so I was able to pull the weights from before. Exactly. They had the wet. Exactly. So you're like the, the patients have documented in their prior visits what is their baseline dry weight. And so when you're when those patients are in the emergency room, the doctor can see that their weight is at this a dry weight is their normal weight. And so the that this is their weight and they're above that. And I need to do this level of treatment based on. There's a lot of other stuff. I'm just, I'm really simple. I'm also, I'm the software engineer. I'm not a clinician. So if any of your like clinical followers say Vince is wrong, you probably would trust that, Yeah. Interesting. So from a doctor perspective, is it like a cardiologist that's using obstructive health or is it like a primary care doctor, is it obstructive health? Our generals using them like our initial ICP, our general users is a primary care clinician, But a lot of clinicians when they're doing the first like 3 minutes of care for a patient, they look at a patient like a primary, like an internal medicine primary care physician before you. Like if you're a surgeon, before you want to like put your patient under for anesthesia and like, let's just say you're going to go get a root canal for

[26:29]like a dentistry that most patients don't want the drill like they you'll go under anesthesia. Anesthesia in itself is a complicated procedure. Dental surgeons go through a number of like medical training and so they access your prior information because if you are on like a blood thinner and then you go under anesthesia, that's a medical complication that they don't that they would be at risk of. And if they know that information of you on a blood. And so it's like clinician just in general want to know the story of the patients. It's like trains in your medical education, your residency, you learn to to story tell. You learn to like figure out what is the differential diagnosis. Meaning like you're problem solving as a doctor, like you, you don't know always what is the exact illness of the patient you're trying to figure out. Like of the information I have, like, what are the like, if you watch house, it's like they write on the whiteboard of like, these are the, like the five illnesses the patient had. And it's never a lupus, but then it might be lupus that one time. And so it's, they're using like prior medical information lab tests. And so the, the, the role that the clinician's doing is to figure out from the information I have, like, what are the potential problems? And then how can I narrow to the best likely case that the information is presenting in front of me to? And so every clinician thinks like that, and that's what we helped solve from like a solution perspective. Can we do like a before and after like before abstractive health like what's the like a patient walks in, the doctor looks at documents like what do

[28:29]they look for and after abstractive health like what? What patient walks in? What do they look for? Like what's the difference between before and after? Like so before the clinician. In both cases they have about 5 minutes of time that they can reasonably spend before they will do care for you. And during that time they will potentially log into the electronic health record their computer system. And if the if they have a front desk staff that is like nagged the patient before the visit to upload their medical record or like call hospitals to get faxes, then they might have that information of outside records in that computer system. If they don't, all they have is specific to that facility, potentially that care that was performed. And so then they may look in that 5 minutes of time, 1 to three recent notes, they'll go to the. And so there's a big recency bias. They'll go what was the most like? Like what was the last thing we talked about? What happened in that visit? And maybe I get a note from a recent hospitalization and I and I read the specific content and then maybe I like to look at a few images and to like, if I'm a radio and then my specialty, I'll look for this specific information. OK, time's up. You then walk in based on the little information you were able to have. And you depend a lot on the patient telling you what's wrong. And so there's a bias that you need the patient to be a good historian to say, like, what are

[30:29]their problems? Gotcha with the. Decency bias with patient bias Exactly a lot of. Biases, a lot of biases. With abstractive, we get about 80% of a patient's longitudinal record, meaning like when they from up to like 20 years ago, wherever they've lived across the country, it pulls in that information. We give a synthesis on their medical history, family history, social. So they're able to in that 5 minute time spent about 3 minutes to review the summary and get like a, a very good high level understanding of the specific care that you've had and what were the key clinical events. They can like dig in quickly to those records, see where it's length and source. And then they can spend the next two minutes potentially like asking further follow up questions like, oh, I see this patient has information on like insomnia. I can ask like what sleep medications are they on? Like, And so you can dig in on like some like specific care wise and then you walk in and then you can talk with the patients specifically about their medical history. Of course, like you still take in the context what they're saying and what they're telling you, but you you know a little bit more about their prior care and so you can do a more informed medical decisions. Like how different is the before and after like from the doctor like like pathway like questions they ask the patient like where they have like never have even concerted asked about the congestive health health, the heart issue, but now because of abstracted health, they thought to ask about it about. You should, yeah, you should. You should think of care on a spectrum so that like if you see 100 patients and you're a primary care physician and like I'm a 40 year old male with like minimal health issues.

[32:35]And so if I go into a visit, maybe my record is not as like like I have like I was mentioning like some insomnia like, and but it's like sometimes when you see a patient, it can be super impactful. And so you should like, it is the instances of that like 5 to 10% that give you pause as a doctor of like this was something that could have been super impactful that I needed to know and that I didn't know. And it makes me take a pause on like, am I doing this care in the best safe manner? So those are the really important. Those are the important cases. And so it's similar it's and then when that happens, you're like, it wasn't the like the typical flu or it wasn't like what I thought it was. There was a little bit more complicated. So and patients want to it's I view it's like if you fly, I'm about to go fly tomorrow to Orlando. Flying. Yeah, you're flying. I'm like gonna fly Delta and like you when you do air travel, you want to not like you want that time when it's like it's delayed and you have that like where it's rainy and it's stormy that like you want to be like the consistency taken care of by the airline, not the 65% of the time when it goes well, that's like abstract. It's like it does the consistency that you will know that the patients like you know their history and like for those very complicated comprehensive cases, you know them well. Gotcha.

[34:35]Oh, very interesting. It's more consistent like. Exactly are. All smoothed over, yeah. Where would edge cases still pop up then? Is it Like? Where would the edge? Cases still I mean that. Abstract that would miss. So we get about 80% of a record and I there's a lot of urgent care facilities that are not connected to the National Health exchange. Like a few years ago, COVID shots were much more meaningful in terms of like knowledge of it. So if you got your COVID shot at City MD, that might not be pulled in too abstractive and so it, it wouldn't have the some of the information. So there's a level when you're making decision as a clinician that you are making your decision based on the information that I have present and you're not often making a medical decision that I have 100% of the information. And so I can say for a fact, I know this is not the case. You're always saying, like, what is true? And so it's like, what COVID shots potentially does the patient have? Or like, does the patient have this prior medication? And so you're collecting information to potentially and hopefully consistently know as much as possible, but it can be much harder to know like what is not true if if that seems intuitive from like a logical level. Does that make sense? Yes, there's yeah, yeah. I'm so like the way I'm picturing right now, there's like up to date open evidence. They're they're like on best practices for medicine.

[36:37]And there's also like the patient specific data that's coming from the bottom like the. Data exactly. Like at what point like this to merge together and there's no need for the human doctor in between? Like where, when, When do you need a human doctor still? So the human clinician is super needed for the like analytical understanding of like what is rights from like a patient understanding perspective on the data. There's still an inherent level like we don't do inference on like what is this diagnosis on the patient chart? And it can be like problematic for what I was saying of you want to potentially say like, what are the list like maybe there's like present in the record these prior diagnosis and you want a human to ask like to judge and to come to an understanding of what the information is. I guess like the best way of looking at it is, again, healthcare has has already have been using staff that are not MD's to assist with the workflow in primary care, like nurse practitioners, physician assistants and like medical students help at like a level in their training and the they're very capable of and doing like a lot of the care. There's decent sites that show that like MD's are good at finding the potentially like very complicated cases like let's say the one to 3% of like when things could go wrong.

[38:41]Gotcha. And I guess I'm in a plain mood of like aviation. And so it's like we have today that like you have autopilot. Yeah, Autoplanes 90. Percent and I and but like when things go wrong, do you still want to have the pilot at the controls to make sure that the plane can safely land? And I think we should view healthcare as the same that we can have like AI, like, just as like computer systems really helped our aviation ecosystem keep and improve like the safety and the job performance of pilots. But we still want them because at the end of the day, the pilot is best to land that plane. And when things go wrong, it's always like the human is best to, to handle the very, very complicated systems. That's, that's where AI is likely to head for the physician. And then separately, I think in the next 5 years, you're going to see a lot of humans not want their communication to be with like AI. They they're going to push back and they want like a human connection to their most sensitive moments that we already don't like when we call like. Let's say United Healthcare and you have like the phone agent and you're trying to just get to like a nurse and you're talking with like an AI system. I think people will want to speak with humans, speak with doctors, speak with someone who can emotionally connect with them about should your like is it time for your loved one, your father to go to Hospice?

[40:45]Are is it is it OK if they come home with you? Like what? Like do you want AI to when you go to the emergency room to tell you that you have leukemia? No. You are human. You want. Human. And so it's like you and so like I and we're, I think you're gonna have a huge push back. Yeah. That humans want clinicians and they want doctors to, to do that, that 30 minute discussion with them because it's such an impactful part of like what they view as health. It's like that, like that human to human connection. Lily, what's the optimal workflow like 10 years? Like what ideal workflow like 10 years later you have abstracted health pulling all the previous data. You have open evidence like coming in and then you like and put the data into open evidence as like what's the best treatment outcome? And then you have the human doctor tell it to the patient. Is that like what the optimal workflow is? Long term, like you have a you have a system that the doctor can understand their patients that whatever that what we're building towards is you're able to get thorough effective intelligence that you can interact on it to build and to communicate. And the clinicians aren't forced to use a system that's not designed for that like traditionally over the like clinicians have not liked their electronic called record computer systems. They've been generally there was huge studies out of 2017 that like were very anti EHR against clinicians and they've been more or less top down from

[42:47]administrative staff for operations like doctors haven't had as much liberty and choice to use the tools that they view best to care for their patients. Yeah yeah yeah. What you're seeing with tools like Open Evidence, with Abstractive Health, yeah, with ambient companies like Freed AI, Heidi, OK. That they can themselves self adopt and like use them and determine how they want to properly treat patients. And they can figure out how to get proper business associate agreements for their practices and to layer on the ecosystem for caring for patients. And then that structure will be designed for clinical intelligence, patient intelligence on the care ecosystem. And then it there will be some sort of interactive layer with billing systems, with scheduling systems, with the administrative and the operational layer. I think you're going to, I think you're going to see a chasm of the clinical AI and the traditional administrative operational AI. And I don't think it's going to be gated as we've seen over the last 15. 20 it'll be 1 cohesive. System, I think it'll be like you'll see a split and you'll see clinical and you'll see operational administrative and there will be communication between the two. But the infrastructure of a clinical AI system is inherently

[44:52]designed on unstructured data that it's doesn't like it takes some context and ontology of like clinical codes and it builds upon that. But it it doesn't necessarily depend on a structured database of storing in like an Excel structure. It can make sense of a vector database and like store for a token based ecosystem. The prior administrative in billing was structured for that prior ontology that isn't AI. And so I think you're going to see that these two ecosystems are designed for different functions and that we're trying to at the moment keep the two together. But I think they're going to divide. And you're seeing the early stage of that with open evidence, with ambience that they exploded outside of the traditional electronic health records. And then there's a push to try to bring it back in. But I think the the chasm has already occurred. And I think you're going to see that there's going to be a like it's going to remain separated at this. Point so for that like flat fee for that one treatment of like just health like how does that how does that transfer from like the billing over to the care? What's like clinicians need to treat and care for patients to do its best for figuring out what is the diagnosis and document on what happened. And then separately they need to submit a claim to get reimbursed for those that collect and to

[46:59]like for insurance with either Medicare Medicaid fee for service insurance or the VA that they filed like billing and claim information both like, like you don't like the clinical ecosystem. You need to save the notes and you need to document it. And you sometimes document it in a way so that you are ensuring you get bills. But the two of them can inherently be separately built that a system that is designed to best treat the patients doesn't need to be the system that is for best for billing. Yeah, that. Makes sense. Treatment and billing are two separate worlds, so you can look at different data and you can look at different different. It's also inherently like there's a risk level when you combine the two, but like you shouldn't be treating a patient just to get reimbursed at a higher rate. It's illegal in a lot of ways to like and there's yeah, but we won't get into all that. That could go on the whole. So longevity is a big trend right now. People all want to live longer. How does like obstructive health fit into that living longer? Picture so we work with some longevity clinicians that are using abstractive health currently and part of the understanding of how to treat a patient and do care for helping a patient live longer is to know them well and to order the right medications to understand like what is their like blood pressure and to monitor it properly. A lot of illnesses in the US are chronic oriented in that if you take care of them earlier and

[49:03]you don't like exacerbate them, you don't end up with later life complications like obesity, diabetes. Another one that I think is going to be up in like studied more. I see the there's like an ear sign on the window that there's early studies that show that hearing loss and tinnitus is correlated with like dementia and Alzheimer's. And so we take for granted in the US like going to concerts or bars that are very much above like 95 decibels and like noise. And we likely by taking control earlier in life of things like your hearing health could have a longevity standpoint of controlling for later things like Alzheimer's, early studies like so stuff like that, that like by knowing your earlier life components and treating them well, that later life components you can have a handle on like with like genetics is another component and blood test, yeah. Very cool. So let's talk about you then. Like the so, so you came to Cornell Tech, the, the, the, the, the Cornell Tech in New York City And then like, how did you decide that this this startup would be a good fit for your experiences, your skill sets? I've been in healthcare 18 years. I studied economics undergraduate at Cornell and I did the Peace Corps actually after at Cornell for three years in Peru. I, when I came back from Peru, I worked at Epic for about two years in the medical records department HIM and then I was a consultant for another seven years with Accenture.

[51:05]And during that time I saw clinicians struggle reviewing and writing clinical summary notes here in New York City. So I was at like Montefiore and NYU and was tasked with deficiency levels on doctors for not writing their summary notes. And so the focus became instead of doing punishment and penalizing doctors for not writing these summary notes, why don't we automate them? And so went back to Cornell, did my research for two years in the space of medical record summarization and spun out obstructive health with my Co founder Giordana to build in the company. We've been at it for four years. And so I'm now on like the Advisory Council of the health information exchange of care quality and deeply knowledgeable of like how the health record ecosystem. And it was like a culmination of all this healthcare background I've been doing. And then I learns the summarization and the AI we'll hear at Cornell. Tap to spin it out. Super. So for for the final question, I always ask the guest, what's the kindest thing anyone's ever done for you? The kindest thing everyone's ever done for you. Anyone. Anyone. Yeah, anyone's done anyone. I would say that. I mean, the kindest person. My wife is my wife, Hannah Ravis. That like one of the hardest things as a startup founder is you have like a roller coaster journey of highs and lows and just like the support and like the rawness of being there for some of my lowest points. And it's like, you know, being

[53:06]just a Good Wife, a good friends and not, not not doing, I told you so being like supportive. That's the thing that comes to mind immediately. That's wonderful. Thanks for sharing. Thank you, Tony.