Ep. #13, Mirroring the ER Brain with AI with Nathan Murray of DocAssistant

Christine Spang: So DocAssistant. It's like an iOS app?

Nathan Murray: Yeah, we have an iOS app. You can use mobile browser as well. The way that doctors use it and mid-levels like NPs and PAs is usually, you know, you'll record the conversation on your phone in the room and then you get back to your computer and then there is where the note is, that's where the clinical recommendations are, and that's where you can interface with the LLM to talk to it about your patient.

Christine: Yeah, I was just wondering, when you sent out the newsletter, were these people then trying to download a new version of the app?

Nathan: Yeah, there was a new version of the app. It's actually pretty funny.

We realized that we were having some transcription issues with an old issue, and we figured out how to fix that.

But then we realized that the setting that we had on that initial app didn't make people automatically update the app, but we learned how to put that in.

But in order to get people to where it auto updates the app, they need to download the new version of the app.

So like 500 users are stuck in the old version until I can manually make them like, "hey, go update this link," they're still stuck on the less accurate transcription model.

But, you know, these are the trials and tribulations.

Isaac Nassimi: Are you guys on React Native or what's your stack?

Nathan: Yeah, we use React Native. Our main companies that we work through are OpenAI, AWS, and then Deepgram, and then we use Whisper as backup and for multilingual for our transcription.

Isaac: Those OTA updates are super cool and super tricky to get right.

I feel like a lot of the mobile world of development is just kind of like crossing your fingers and being like, "this fast lane script usually works, and I hope it works this time."

Nathan: Yeah, there's a very tight communication loop between users. And then our engineers and me in the middle being like, "Pietro and Josh, this guy's having this problem right now. Can you fix it?"

Isaac: And I don't want to get too tactical on the software development side, but I'm curious, are you guys able to get good telemetry for debugging issues and things like that?

Nathan: What do you mean by good telemetry?

Isaac: Well, I mean, I think you're kind of already explaining it in that, okay, a user reports a bug. Do the devs go like, "hey, can they say more?" Or do they go, "oh, I'll track it down. I got it."

Nathan: It's a combination of both.

I usually just create a group text between the user, me and them.

Because they often will be like, "hey, can you ask them to do this?" And so I found that it's just more efficient if I can loop them directly in and then they'll have the user try certain things and oftentimes we'll find that user error that was causing the bug.

Like, we had one the other day where our main engineer was working on this thing for, like, three hours. She's like, "I cannot figure out why they can't log in."

And then finally we figured out that she had saved her username with an extra L in the email.

So she was, like, able to log in on, like, mobile browser, but not the app and we couldn't figure out why.

And he's like, "I think the password's wrong." She showed the password and the password was right.

And then finally after three hours, it's like, "wait a second, I see the issue. She put an extra L in her username that auto saved."

So there was no real bug, it was just an incorrectly saved username. But yeah, that's the fun stuff that you get to deal with.

Isaac: It's always user input error.

Nathan: He was so mad. He's just like "that was three hours of my life."

Christine: Yeah, I come from the UNIX neckbeard world where the kind of gruff dude on the other end of that would just say, a "PEBKAC." problem exists between keyboard and computer."

I don't really recommend saying that to anybody these days. It's kind of a jerk move.

Nathan: Yeah, I haven't heard that one. I've learned a lot of new words, but that's not in the vocabulary yet.

Christine: Yeah, the thing that I'm most excited about, actually about this conversation is that I feel like you're inhabiting a totally different world that I know nothing about and I have a lot of questions.

Nathan: Yeah, for sure. Maybe it's helpful if I introduce myself.

I'm Nathan Murray. I'm an ER physician and I founded DocAssistant. It's an integrated AI scribe, clinical decision support tool.

So we started building it probably about nine or ten months ago now.

So I've just started dipping my toes in the tech world, learning about tech stacks, AI and all these things. It's been a great experience.

My co-founders are both very deeply technical, have started multiple companies, worked at all the big companies themselves. So it's been a cool experience.

And, yeah, thanks for having me today.

Isaac: Thanks for coming.

I'm curious what made you make the leap, right? Because you're not technical and you have an intense amount of context on the medical side of things.

What gave you that confidence to be able to jump into the software side?

Nathan:

I just felt like there weren't any good AI tools that guided clinical decision making in the emergency medicine environment. You have ChatGPT, you have all these LLMs, but there weren't tools that were specifically structured to fit naturally into our workflow in a meaningful way. And so it felt like there should be.

And so it just felt like a cool problem to try to solve. And so I started working with my friend Pietro Sette on a solution for it back in October of 2024.

So we started kind of just building together. He's an absolute tech wizard.

So, I was able to kind of lean into him and be like, "hey, this is the thing that I think would work, or this is like a design that I think would fit into my clinical workflow. You know, it feels like there's this very powerful, super intelligent AI and then we have our kind of archaic medicine, and it's like there's just not good glue combining the two."

At least in the emergency medicine setting. I can't really speak for other specialties as well.

And so I just felt like if we could build a solution that helped glue them together. It would be like a very useful thing that could improve patient outcomes worldwide.

So we started building in October. We created a product that essentially just thought like an ED doctor.

Like, essentially we were just trying to mirror the AI brain to match the ER brain.

So the way the MVP worked is a user would come back after seeing a patient and type in information about that patient.

And then the AI would generate a differential diagnosis, testing recommendations, treatment recommendations, and then next steps.

And that's how you think as an ED doctor. You see a chief complaint, you start thinking, do differential.

You talk to the patient, you get a history, you do a physical exam, and during those you're narrowing your differential.

And then you're like, all right, these additional tests, I can order a CT scan, an EKG, basic labs, whatever it is that'll help further narrow my differential.

And then you try to get to a diagnosis. Sometimes you can't, but you try to.

And you're stabilizing the patient at the same time kind of in parallel.

And then once you've gotten to like stabilization or diagnosis, then you determine if you're going to admit the patient or disposition the patient.

So we built the AI to do exactly that. And then things that doctors really like are citations.

Everybody likes to be able to check their sources. Otherwise it's like the AI, uh, is just this kind of scary black box.

So we partnered with a company called StatPearls. They're one of the biggest content companies in medicine.

So they have over 10,000 peer reviewed published medical articles. And then we integrated all of their articles into our AI.

And so now the AI, when it generates these recommendations, it'll actually link the article from which it was able to kind of source that information from, so the user can click on the article and read more about whatever the disease is, the medication, whatever the AI is recommending.

People really like that, the citation aspect helped to build confidence in the product.

But what we found after that was that it was hard for users in a busy ER environment to sit down and type out all the information about their patient. So we realized that what we needed to do is layer in an AI scribe.

And I don't know if you guys are familiar with AI scribes, they've kind of blown up in the medical industry over the last six to 12 months.

But essentially what it is, is using real language processing, like a live transcription engine will capture a conversation and then from that write a patient note.

What we realized is what our real mission is to try to get evidence based clinical recommendations throughout the world.

And so we realized if we layer an AI scribe on, users don't have to come back and type the information, they'll actually be able to just have the information generated for them by the time they get back to their desk, in addition to also having their patient note written for them.

So then we built that aspect to it that's been an entirely different lift with a whole different set of problems, but it's been really great and we made a lot of progress, especially recently.

And that's actually made it significantly more sticky. That's how we've gotten a lot more user retention lately.

And then even beyond that-- So I was presenting at a conference, I was just kind of pitching at a booth and I was just trying to sell the product.

A medical director and me started to talk and I was like, "hey, do you guys want to trial our AI scribe clinical decision support tool?"

He's like, "we're not really looking for that right now, but if it could help us with filling or with reimbursement at the level of the clinical note, I'd be interested."

I was like, "that makes a lot of sense."

And so we started building a chart level analyzer, something that reads the chart that the clinician wrote and that the AI wrote together, analyzes what level it is right now, what level it potentially could be based on how sick the patient is and what information is potentially missing that could help to optimize reimbursement.

And what we found is that has been kind of the glue that has allowed us to get user retention and more importantly medical director attention that we can kind of backpack the clinical recommendations on.

Because a lot of clinicians, maybe they're busy, they have their own kind of patterns, they don't really care so much for the clinical recommendations, even though I think it's really important that everyone kind of double checks their work when we have the super intelligent AI that can, by putting that, backpacking it on an AI scribe that helps people to not write notes which they love, and then also optimizing reimbursement which the decision makers, like medical directors, love.

We've kind of been able to bundle this whole product like that actually is getting more traction and getting people to use like a clinical decision support tool.

Christine: I'm going to have a lot of really dumb questions here because I don't know the medical world at all.

First off, what's a differential?

Nathan: So a differential is when you don't know what the disease is, but somebody has chest pain, it could be one of 10 things.

The differential is those 10 things. So it's like it could be a heart attack, it could be a pneumonia, it could be a blood clot.

Christine: Okay, it's like your possibility space. Like, what are the possible causes of these symptoms?

Nathan: Yeah, short for differential diagnosis.

Christine: Okay. And I'd love to just get a little bit more color on what life is like in emergency medicine? Like, pre-Doc Assistant, what was your life like when you were just on the floor?

Nathan: Yeah, it's busy. It depends on the ER you're working in.

Some are higher acuity, patients are more sick, some are lower acuity, patients are less sick.

But in residency, I'd walk in usually around 6:00am. 6:00am in the ER is when you get the CHF exacerbations.

You get a cortisol spike in patients and all of a sudden they can't breathe.

So they come in usually between 6 and 8, we'd have one or two severe exacerbations.

People really short of breath, their lungs are filled with fluid. Those are the sick ones.

The sick ones for me are the fun ones because you get to use your skills, both your diagnostic skills and your procedural skills.

So maybe you'll have to intubate that patient to protect their airway because they're desaturating.

Maybe you have to figure out is it actually a CHF exhibition. Do they have a PE instead? Do I need to get a CT angio? Do they have some weird other disease?

So you have a few critical patients like that, and then usually around 10am you start to get slammed. That's when the people wake up, they have their nice cup of coffee, they take a shower, and they go, I'm going to go to the ER now.

And so then you see all of them. You know, cut on a finger, belly pain for four days, back pain for six years, all the things that may or may not need to be in the ER.

And then usually ERs are pretty busy. On average, nationally, you see about 2.3 patients per hour.

So you'll see anything under the sun from a heart attack. Like we talked about, a CHF exacerbation, belly pain, they'll do a laceration repair.

You might get a really bad car accident or a gunshot wound.

Our job in the ER isn't to be an expert at any one thing other than stabilization. Our job is to stabilize and disposition. So try to figure out what's going on the best you can. Make sure you don't hurt anyone. Make sure that you stabilize to the best of your ability. And then get them to where they need to go, whether that's the ICU, the floors in the hospital, or sending them home with follow up.

Christine: Okay. So there's like kind of a lot going on. You're seeing patients with all sorts of different issues.

Sounds like you're on your feet and you're kind of talking and interacting with different people. And there's like a lot of different tasks that you're juggling.

Nathan: Yeah, it's all about triage and task juggling because there's at any given time there's like 20 things to do.

So you just have to constantly be asking yourself, is this the most important thing that I could be doing right now?

Which has actually been really helpful as I transition to startup mode.

It's just like, you know, every day my task list is about 20 and I'm like, what are the three most important? And I keep re-asking myself that every hour or two.

Christine: Were there like any tools available on the floor that were technology based tools that were helping you out or was it just like pen and paper?

Nathan: Yeah, there's a website called Wik EM. It's actually an open source, really great product where it's similar to Wikipedia but for emergency medicine, where it has like every emergency medicine disease.

And it says like, what differential diagnoses you should be thinking about for this chief complaint, what the workup should be.

Christine: Okay. And are you searching for stuff on this, like on a computer in the room or like, what is it like?

Nathan: Yeah, so it's open on a separate tab. You go back to your computer and you see that tab.

Christine: Okay. You have like a laptop.

Nathan: No, you have your own like desktop.

Christine: Okay. A desktop machine. Okay.

Isaac: As a hypochondriac, I'm unhappy that you just gave me this knowledge that this site exists.

Nathan: Oh, don't worry. It's much harder to use than just like googling your problem and then being scared by the Gemini response or whatever it is.

Isaac: Okay, thank you.

Christine: Okay, so while you're walking around, you have a question, you go walk over to this desktop and you type in like some search query?

Nathan: Yeah, it was more helpful in residency when I was like learning how to see patients and stuff.

Now I kind of just use it to confirm my test that I'm ordering.

But the cool thing is we actually wired it into DocAssistant too.

So like now DocAssistant will actually source the recommendations and then link the WikEM articles for me.

And so, rather than me going and finding the article, it's actually just brought directly to me and I can just click on it from DocAssistant and it brings me to the page.

Christine: Okay, so it sounds like having had this experience, you know, working with these tools, they're kind of clunky. There's a lot going on.

You guys had the insight that, "what if you could have more modern tools accessible on your phone while you're doing this?" Is that a right kind of synopsis?

Nathan: Actually we still use it mainly on the computer, in practice.

The phone is really useful for capturing the conversation, but that's not where the information gathering occurs.

That's like not where like the bandwidth funnel is.

It's still on the computer because by the time you get back to your computer, you need to write the note or have DocAssistant write the note or whatever service you're using.

But then DocAssistant is great because it's like we have a tab for the patient note and then we have a tab for the clinical recommendations.

Christine: Okay. So it's really more about the interface and just like upgrading that experience of working with the tool than necessarily like moving what device you're accessing it from?

Nathan: Yeah.

I do think the direction this is going to go, and this is on my two-year plan, is I think that there's going to be a world in 12 to 24 months where people start to wear smart devices, whether it be like Meta glasses or some sort of AirPod where there starts to be a bi-directional interaction between the AI and the clinician to where you can populate information for you, live in the room.

So right now the way it works is you have an app on your phone, it records the conversation, it sends it back, processes through the AI, outputs the patient note, the computer and all of that.

But what you could do is if you had a smart device, is you could then cue the AI to then push that information back up to you live.

So maybe on the glasses it populates a differential diagnosis as you're seeing the patient or talking to the patient.

Or if you're having smart glasses, you could record a rash, it could process through the AI.

Rashes are very hard. There's so many different rashes that all look similar.

So having like that live AI feedback where it could be like, "this looks like an allergic reaction" or "this looks like a drug rash," it would help with the conversation with the patient.

Or it could even prompt you like, "hey, you forgot to ask about this important question with the patient."

So that's kind of what we're working towards. That's kind of my goal and like, kind of my vision is like, I'd love to create...

We have this super intelligent AI that we haven't figured out how to actually use at the application level appropriately in medicine yet. How do we get there? And I think that's one of the next steps is: get a deep integration with the EHR and then get a bi-directional live feedback loop between the clinician and the AI to truly augment decision making and completely offload all administrative tasks.

Christine: Yeah, I think what you're doing right now is super cool by the way.

Like basically, all of this cutting edge tech, all of the biggest value that is going to happen is in people taking that , being subject matter domain experts, and applying it to what are the workflows and what are the needs of this particular segment of people who are trying to get something done.

And you know, that's exactly what you guys are doing with DocAssistant.

You mentioned before that like when you were in the ER, in the emergency room, getting started, you saw MedTech as being this really baroque thing and obviously you saw an opportunity to upgrade that.

Have there been any particular challenges towards actually building tech that is adopted by folks in the field?

And what have been the particular challenges that you guys have been able to solve in order to get uptake?

Nathan: Yeah, a ton.

So the biggest thing is just fitting it into the actual workflow because you can build a really cool tool that you think is really useful, but if people don't want to bring it into their busy schedule or their busy workflow, then they're not going to use it.

And that was the biggest problem that we found when we first started. We built this tool that I thought was really cool.

You know, it brought WikEM and StatPearls links directly to the clinician based on what the patient had. I thought it was super useful, but then our retention wasn't very good.

People just weren't using it. And the reason when you talk to users is because it's like ,"hey, this is cool, I'm really busy, I don't have time to look it up,"you know, all these things.

And some clinicians liked it more than others, some would use it more than others.

But that was the biggest thing was just like it's just not saving me time. And at the end of the day people protect their time more than anything else.

So yeah, so the way we solved for that one was layering on the AI scribe and then also making it stickier with allowing for increased reimbursement levels, that makes people really want to use the tool because, "hey, I can save time and make more money."

And then we kind of just package the clinical recommendations with that. That's the biggest hurdle that we've run into is just making it actually useful and actually add value and save time rather than just being a cool thing on paper.

And it probably would improve patient outcomes if people used it. But if they don't want to use it, then it's useless.

Christine: Yeah. Who's the buyer in this situation? Is it the hospital? Is it the doctors?

Nathan: Yeah, it's kind of a two-tier system. So our primary, ideal customer is a medical director.

For us, it makes a lot more sense for two reasons. One, a lot of emergency departments are going to require top down approval.

There's more red tape in some ERs, less red tape in others. But in the ones that have more red tape, like we're currently doing a pilot with um, ET Houston, which is obviously like a big institution, very academic, they have a lot of red tape.

We need medical director approval and she needs to push it through her ID department. She has to get approval from a bunch of committees. It's a long sales cycle, all of these things.

She's our ideal customer because when she likes it and she pushes it, she's going to distribute it to 50 to 100 users all at once.

And she kind of acts as a moat too that has protected that ER from adopting other AI scribes or competitors because it's like there's so much red tape that it's so hard to push and cut through that if she does like us and she does do it, it's a moat and a distribution network.

So she's our ideal customer. The medical director is our ideal customer.

But we still do B2C as well. Anyone can sign up, do a free trial and get using it immediately.

And obviously there's a lot less friction with that sales cycle and it's a lot quicker but it's less protected and it's a less reproducible distribution network.

A lot of the times they'll use it for a little bit and then their ER will start trialing a different AI scribe.

But still, we've had users that brought it to their medical director, like "hey, I'm using this one, can we use this?"

And so we've kind of focused more on the B2B approach, but we're not ignoring the B2C approach and it's still an important thing for us.

Isaac: I think that's really powerful. I mean that's as far as I know, how Slack really took off was kind of having this dual model of creating these little cells inside of companies that organically started using their product and then would kind of push it wider or go up the chain of command to get some buy in.

And then also some top-down adoption as well, especially once they created some brand equity and some brand value.

I really like what you guys are focusing on as well because I think there's a lot of people saying like, "hey, you know, in three years the world's going to look like this or it's going to look like this."

And there are always different ideas and generally we don't know. And we've seen a lot of AI features come out recently that don't really get adopted.

But what does seem to get adopted are passive features, things that don't require the users to actually go learn how to do something or do something. It's just there for them.

And like for example, the AI scribe, awesome. I can't even imagine the life of a doctor where you're seeing 2.5 patients an hour.

If you don't find time to sit down and write down what happened and actually log everything well, you're going to lose 80% of that information once you context switch to the next patient.

So, you know, you don't just get time back, you actually get accuracy back and effort and this mental load and you can just context drop the last patient as soon as you're done.

Nathan: Totally.

Isaac: Where do you think-- In that model though, you've got these passive features that you're pushing or that people are actually pulling on and you've got some active features, right, like things like helping with diagnosis and whatnot.

Where do you think you're going from there? Where do you think the next 12 months looks like for you guys, feature wise, and how have you been able to vet that?

Nathan: So the next step is going to be integrating into EHRs.

The way that you do that is once you get an institutional contract , you figure out who their EHR provider is, and then you reach out to their rep within that EHR company.

That's the next step because that'll kind of embed us more within the EHR system where the note automatically will upload into the EHR.

So that's kind of the next six months is kind of focus on EHR integration, just keep getting as many emergency departments, as many clinicians using DocAssistant as we can.

And for me, what I really want to do is I want to build a product that utilizes, like I talked about, AI technology to its fullest extent.

So what I really want to do is I want to be able to create some pilots with some of these ERs, where we start to play with different ways of getting the note to automatically update into the EHR without the clinician having to open up their phone and click record.

There's a few ways we can do that. You can do an ambient listening device. Like you could just have your phone be ambient listening, or you could wear an ambient listening device and then you could use keyword triggers to be like, "hey, I'm seeing a patient in room 13. Start note now," and then you can end it at the end.

That's a little clunkier, but that's one potential solution where they could walk into the room, not touch anything, start the recording, end the recording, push to the EHR, which is kind of what I'm seeing.

A cooler thing that I want to do, but we haven't gotten to this point yet, would be geotagging an ER to where the phone walks past a little sensor in the room and it automatically starts recording for that room.

I think that would be a better solution. And actually it would probably be a more technologically advanced solution that we could actually bundle as a company and then resell, rather than just like ambient listening.

Like, this is a proprietary geotagging DocAssistant system that we could take. We could not only use it in different ERs, only do expand one day to the inpatient side clinics, all of those things, you could reproduce it there as well.

So that really excites me is like the idea of taking out the additional pain points. The pain points right now of using an AI scribe, you have to pull it up on your phone. You have to make sure the microphone's capturing closely or well, all these different things.

So removing those extra steps one by one to where it just becomes this true integrated technology.

The whole idea was it's a AI doctor assistant on your shoulder, "DocAssistant" that helps guide you with clinical decision making and does all of your administrative tasks for you, so it allows you to focus on the patient and be a better doctor.

So that's kind of what I see is first, first EHR integration, figure out how to do a touchless system, uh, maybe ambient listening, maybe geotagging.

And then after that the smart tech, Meta glasses, things like that with bi-directional, I guess it would really be a three way communication between the EHR, you could pull from previous records, the AI, which is acting as a smart agent to go pull that information, feed more information to it, and then the doctor who's the primary information gatherer, asking all the right questions, talking to the patient, doing the physical exam, doing whatever procedures need to be done.

And then getting live feedback from the AI that pulled from the EHR to make better decisions, have a better conversation, et cetera, et cetera.

Isaac: Wow, your roadmap is better defined than most startups, I would say. Congratulations.

It really seems to be an unfair advantage for you, right? Having worked in that position, having been your own customer essentially, knowing what would work, what wouldn't, all that kind of stuff.

So clear advantage in coming in and working at a startup in tech.

What has been difficult to adapt to, what's been difficult to learn and really wrap your head around in this change?

Nathan: I guess the biggest thing that I've had to focus on is learning the language of the engineers since I didn't have any technical background.

And I've always loved computers and stuff, but I've just never, I don't know, I just always was kind of down the medicine route.

So learning, you know, their terminology, figuring, you know, they're busy, they're, they're, they work hard.

And so like I try to have to sit down and make them like, "hey, teach me about this thing that you're doing, teach me about what this process is, teach me about what batch processing is, or teach me about this thing."

It's been very important to like improve the quality of the iteration loops.

Like what I found at first was I would get feedback from users, I would talk to them, there would be a bit of a disconnect like where they wouldn't understand what I was saying medically and then I wouldn't understand what they were saying from the tech perspective.

So I've tried to learn a little bit more about coding. I've learned a lot more about AI and all those things are not necessarily because I'm doing it myself, but because it improves the speed and the quality of the iteration loops between the users, me and them.

So that's been the biggest hurdles so far. And then also just startups in general. It's been a really cool, fun learning experience.

The biggest lesson that I learned is that at the very beginning I tried to over engineer everything because I wasn't an engineer by trade or by training.

So I had this big idea, like, okay, cool, we're going to build this system, it's going to have all these different parts, the AI is going to work like this and it's going to do this for the doctor and then you created that.

They spent a bunch of time doing that, the engineers. And then users are like, "this is clunky and not working."

And then slowly we stripped away 1, 2, 3, 4, 5. And it's just like at the end of it it's like, "oh, I literally just needed the one piece. And I made them do all this extra work. I over engineered everything. It didn't work and then we scrapped it all and now it works."

So starting small and then iterating based on user feedback rather than being like, oh, I know the solution and then creating this big clunky thing that no one actually likes has been the biggest lesson for me.

Isaac: That last one is awesome and I think it takes people too long or some people never learn.

I always say to the product team, ship the smallest discrete unit of product possible.

And yeah, it's not going to be as awesome as it could be. And you'll feel like it's not really getting like a fair shake if you're shipping less.

But really, if your feature can't work with a smaller subset of the features, then it's not going to work, period.

So that's a really awesome takeaway.

Right now you're kind of load balancing between your previous job and kind of your new one. Right?

Can I hear a little more about that?

Nathan: Yeah. So today is my last shift working these hours.

So pretty much for the beginning, for the first little bit, I had to bootstrap everything. I had to convince my engineers to take on the product at a really high level and whatnot.

Once we built an MVP, you know, got some traction, worked together for about four months, they did join me as co-founders which helped reduce the financial load.

But then we started to kind of ramp up GTM and stuff and you know, I was kind of stuck in this limbo period.

I'm a first-time founder. It's not going to be super easy to go get fundraising. We need to show customers, we need to show users. So what I did is I just took on a really heavy workload.

For the last four months I've been working about 70 hours a week clinically and still doing this, you know, trying to talk to users and engineers, do everything.

So it's been a pretty crazy lift for the last four months but now we've just finally started to get traction on the fundraising side and now I'm transitioning from working a lot of crazy hours.

Tonight it's my last shift. This is my 20th shift in 24 days. 12 hour shifts all overnight so that I could do meetings during the day.

I'll now be working like one shift per week. Because it's still important for me to practice clinically and understand the product and understand our iterations on the product and whatnot.

But, but I'll be able to focus more fully on DocAssistant, which will be great.

Isaac: That's awesome. That is like the true hustle.

I don't think I've ever heard of a founder hustling that hard because you know, a lot of people work at their 40 hour shifts, or 40 hour work weeks and then you know, put in the night time on the startup.

You're doing the actual opposite and also 70 hour weeks.

Nathan: Yeah, it's definitely been really cool. It's been really rewarding. I'm proud of myself but it's not sustainable.

So I'm excited to be transitioning to a healthier work--it's not even work-life balance, I'll still be working on it, but it'll be like a healthier sleep schedule.

Christine: I think it's really cool that all of I would say the hardest working people I know have like some sort of medical training background because sitting in front of a computer just really can't compare. Like you guys go through the ringer.

I also think it's like really funny that we've been talking about like "oh, you know, I'm uh, not technical, etc."

And like you're a fucking doctor and like, you know, software engineers, I think we're just trading different buckets of jargon.

You guys got your own set of technical things, like the human body is pretty complicated.

It's just really inspiring to see that, it seems like you did a huge amount of training in this area and really gained this domain expertise.

And you pretty fearlessly jumped into a whole different world when you saw that there was this opportunity.

And also the timing is, I think, really, really great right now just for folks that are grabbing onto opportunities like this.

And I'm super excited to see where DocAssistant goes.

Nathan: Thanks. Yeah, it's an exciting time.

You know, it's like the A.I. revolution. So it's fun to kind of be able to try to bring that value to people and try to improve patient outcomes at the end of the day is the goal.

Isaac: Awesome. I think we're nearly at time and first we do a section called Picks where we just bring something cool that we've found recently that we're excited about or obsessed with or whatever.

It doesn't have to be anything like tech related. It can be whatever. And a good example is last week, I think Spang's pick was a floss. It was like this type of smart floss. What is it called? I actually forget the name.

Christine: Dr. Tungs. I just ordered a 10-pack on Amazon last night. I was running low.

Isaac: I bought some and it's like, I wouldn't call it life changing, but it's dental hygiene changing for sure.

Christine: Good stuff.

Isaac: It's crazy how much comes out, you know?

Anyways, so my pick for this week is actually, it's kind of general and a little odd, but it's having a skincare routine.

And I feel like as a guy, you generally don't talk about skincare routine. You're like, real proud of yourself if you put moisturizer and like sunscreen on occasionally if you're gonna go in the sun.

But even just some basic stuff and, you know, trading with ChatGPT and learning a little bit about it, I feel like I've seen just like a massive improvement in a matter of a few months.

So that's my pick for the week.

Nathan: Nice. I actually just tried to start my skincare routine as well. I attempted to about three or four months ago, and then I kind of just let it fall by the wayside and haven't really been doing it.

But I totally agree with you. It's important and I mean to do it. And I have all of the, my girlfriend calls them "potions" in my little kit, but I just haven't been putting them on.

Christine: Well, maybe once you're not working two intense jobs at the same time, you'll be able to take care of your skin.

Nathan: Yeah. Tomorrow is the real beginning of my skincare journey.

Isaac: Yeah, no excuse after that.

Nathan: No excuse.

Christine: This is really, really funny, Isaac, because in honor of Summer, I brought a sunscreen moisturizer for my picks this week.

Isaac: No way.

Christine: And I use this stuff, SPF 30 for my face. It's mineral sunscreen moisturizer. But then I have a friend of a friend who's like, incredibly sensitive.

She really doesn't like weird textures and stuff. And this stuff, Skin Aqua, is like a Japanese sunscreen moisturizer that really doesn't feel sticky at all.

So I've been enjoying this for my arms when I go outside. I hope it doesn't, like, kill coral or anything.

I cannot read the ingredients. It's all Japanese.

Nathan: Should I grab something for a pick? Is that kind of the next?

Christine: Yeah, what are you psyched about right now, Nate?

Nathan: I guess it would be skincare too. I literally was, like, excited to start this nightly process as well. It was the one new thing in my life, I guess.

Christine: What's the anchor for your skincare routine? Is there a product?

Nathan: Let me see. I was told that this is the best brand. Ordinary.

Isaac: Oh, Ordinary. I had that exact thing. Yeah.

Christine: Oh, my God.

Nathan: Yeah. I also had a conversation with ChatGPT, and it recommended Ordinary. So maybe if Ordinary is in cahoots with Sam Altman, they were smart to do that because they got both of us using their product.

But, yeah, I guess it has, this is for the bags under the eyes, which I've always had an issue with ever since residency.

And then this one's the Hyaluronic Acid, I guess.

I guess it's supposed to make me glow and look younger. So I'm excited to try this off. I just gotta actually do it.

Christine: I keep saying that the 20th century culture was driven by television. And now in the 21st century, culture is gonna be driven by talking to AI.

And I feel like you guys have just proven that right now.

Isaac: Yeah. It's gonna be starting trends.

You know, there's all this stuff that cross up. Like, you hear about Labubu, if you guys know what that is, just comes out of nowhere, there's no IP behind it, and suddenly everyone's putting these little charms on their purses and stuff.

Christine: I think the next two years is going to be quite wild because, you know, there's no SEO for LLMs yet.

And so, everybody who figures it out the fastest is the ones who are getting the recommendations.

Nathan: And we had a meeting about that yesterday.

Christine: It's a weird world.

Nathan: Yeah. Like, how do we maximize getting referenced by ChatGPT? Because that's truly going to be the exposure.

Christine: Yeah. Where it's like, there's no, like, page rank yet.

Nathan: Yeah.

Christine: But they're pulling stuff.

Isaac: No one knows how it's going to get gamed, but everyone knows that it's going to get gamed in a very dirty way for a while until it gets figured out.

Christine: It's probably a couple of years until we have some framework.

Nathan: Yeah, totally. Cool, guys. Well, thanks so much for having me on. It was a fun conversation.

Isaac: Well, thank you for coming.

Christine: Yeah. Really great to have you here, Nate, and best of luck.

Nathan: Thanks, guys.