AIMinds #006 | Fern Cowan, Head of Growth at DeepCura
About this episode
Fern Cowan is the Head of Growth at DeepCura AI. He compares his experiences and skills to a recipe; with the resources at hand, he creates the best possible outcome. Fueled by this perspective, Fern transitioned from beta production to coding, an unexpected turn influenced by the decline of the advertising industry. In the midst of this transition, he dabbled in trading, adding another layer to his diverse skill set. His journey portrays his adaptation to changes in market trends, always utilizing his unique knowledge inventory to forge ahead.
Listen to the episode on Spotify, Apple Podcast, Podcast addicts, Castbox. You can also watch this episode on YouTube.
In this episode of AIMinds, we explore the world of AI medical scribing with Fern Cowan. Transitioning from culinary arts to videography to launching a startup, Fern shares his journey and the inspiration behind DeepCura. He talks about the challenges of breaking into the healthcare industry and the innovative solutions DeepCura offers. Check out the full conversation as they tackle the intersection of technology and healthcare, the power of AI in redefining industries, and the motivation behind overcoming obstacles.
Here are some highlights from this great episode:
The Birth of DeepCura: Learn about the unique inspiration behind DeepCura and how Fern identified the opportunity to revolutionize healthcare automation at a time when the advertising industry was struggling.
Navigating Healthcare Regulations: Fern discusses the challenges of breaking into the healthcare industry, emphasizing the importance of HIPAA compliance and FDA regulations, and how DeepCura has innovatively addressed these concerns.
AI-Assisted Diagnosis & Clinical Workflows: Discover how DeepCura's platform offers AI-assisted diagnosis and enables physicians to create customizable, detailed notes and clinical prompts tailored to their individual needs.
The Advantage of Being Agile: As a small team, Fern explains how DeepCura leverages the power of AI to swiftly innovate and stay ahead of industry changes, making it a formidable force despite competing with larger, well-funded organizations.
Fun Fact: DeepCura's AI engine allows doctors to load prompts with a single click, enabling the efficient processing of patient conversations recorded during appointments.
Show Notes:
00:00 Transition from trading to coding via AI
04:50 AI medical startup automation
09:52 Advanced clinical note customization and automation explained
13:36 Small team, big impact in AI innovation
15:47 Users trust speedy updates, crucial in healthcare
18:41 DeepCura offers transparent, doctor-customizable AI solutions
24:21 Workflow involves recording, loading prompts, and transcribing
25:18 Different thread architecture in healthcare for GPT
28:20 Wrap up
More Quotes from Fern:
Transcript:
Demetrios:
Basically, you transitioned from the culinary arts into videography. Then, by way of San Francisco, you got inspired, probably by all of the software engineers that you were finding yourself around. And then you went to Stanford for a little bit, and you said, you know what, I can code. I'm going to go and do this. With help of chat, GPT, and upwork, you've been able to create DeepCura. And I am really excited to hear what exactly the inspiration behind this was. Why did you decide to create DeepCura, and what is the pain that you're trying to solve, for sure?
Fern Cowan:
Absolutely. So the way I think about it is, I think in my head, I have an inventory, right? Like a knowledge inventory. And now that I think about my past is the pattern that I never really think about it. I think this in some way was intuitive. But now that I think backwards, in hindsight, it's more like I was trying to build a web based on the inventory of my knowledge and based on what I had. It was like using sort of like this recipe, analogy recipe, right? I had these ingredients, so I'm going to make this dish, right? So that's exactly what I've been doing. And the reason why, DeepCura, it was just the combination of the market, meaning that the advertising industry was dead. And then also before, I had, like, a little between the transition from beta production to coding, I did a little bit of trading.
Fern Cowan:
Everyone was a trader with cryptocurrencies, so I did a little bit of that, and I learned about markets, like, how certain industries perform better in a recession. So I just know that the healthcare industry is like a defensive market, right? So my hypothesis was, you know what? Now that things are really down, healthcare is always, like, a good industry, right? Like, even if it's in a recession. That was my hypothesis. So I thought, you know what? Let's transition. Let's do coding. And what really gave me the confidence of transitioning this industry was the combination of being working with AI, medical stripe since 2018, and also the fact that Chat GPT, and AI was so powerful to code. So if you're creative enough and you have a good understanding of concepts, or at least the architecture, you can give that to Chat GPT, and obviously it will give you a code full of bugs, but if you work hard enough, you can solve those bugs on your own and then just move to the next step. And really, the way that I've been handling the roadblocks has been through chat, GPT, and upwork, and pretty much that has been my strategy.
Fern Cowan:
Now, of course, the greatest part is that every time that I get into a roadblock, I learn from it, and that gets permanently stored in my mind. Right. So basically, through the DeepCura, it's paying my knowledge and my experience. Like, instead of going to university, I'm being paid to learn, and I think that's the greatest hack I've been able to accidentally learn in the industry.
Demetrios:
Well, I have to say, you are one of the lucky ones that can learn something and then remember it. I learn something, and then I forget it about 20 minutes after I learn it. So that is a nice skill to have, or a blessing, I guess, that you have been blessed with. And me, not so much.
Fern Cowan:
Yeah. Sorry that interrupt you, but let me be on that for the people. I'm exactly the same like you, but I feel like I only remember the things that I feel passionate about, and that's the only reason why I remember if I wouldn't care, I have forgotten everything that I have learned in high school so far. You cannot ask me about algebra or anything like that. Yeah, I guess I just wanted to say that to be fair, too, as well.
Demetrios:
Yeah, it makes sense. It gives you a lot more of an attachment to something if you are passionate about it and you're learning about it, and that passion shines through and makes sure that it gets stored in the memory bank somewhere.
Fern Cowan:
Yeah, absolutely.
Demetrios:
So talk to me about what DeepCura does.
Fern Cowan:
Sure. Yeah. So we are an AI medical strike company, and we basically automate the workflows of physicians. It can also be the workflows of hospitals, but mainly right now, the company attracts mostly individual providers and small practices. The reason being is we're a young company, so it's very hard for a hospital and say, oh, yeah, let's go with this nine month old startup. Right. The best part of this is that it's like momentum, right? Once you conquer the small niche, do you know that the boys pass around, and then the hospitals are like, oh, they're doing this. And then they evaluate their current solution.
Fern Cowan:
They're like, wait a second, our current solution doesn't can do, like, computer vision. So then they get curious, and then they will have to pressure their current solution or simply just give up and say, okay, let's at least analyze your aws cloud to see if you're actually HIPAA compliant, and then we can talk, which obviously we are open to because we are actually 100% HIPAA compliant. But I think that's what I imagine it will be their perspective. But it's really hard. It's really hard to sell to a hospital for the same reason.
Demetrios:
Knowing your background and knowing that you jumped in headfirst into the healthcare space, how did you learn what the workflows were that you needed to automate?
Fern Cowan:
Sure. Yeah. So something. A very popular misconception that people get in the healthcare industry is that you need a doctor in your team or the doctor needs to be a co founder. And I'm going to tell you why that is not the case. The reason why that is not the case because when you are serving physicians, you have 100 doctors that are actually giving you direct feedback on real time, and their feedback is way better. The reason why is because they're actually paying you to solve their problem. So they're going to give you harsh feedback.
Fern Cowan:
They're not going to be polite. They're going to tell you, you know what, this sucks, or you have to do this, or we would like this feature. And then also you can add on top of that, you can be more aggressive and also ask them, be active about it. And then you actually get better feedback than having a doctor co founder because you have like 100 sources of clinical feedback. So you actually gather an incredible amount of very rich knowledge from different specialties. So you become really smart if you really take that approach. So I just wanted to throw that in because it's something that I get pretty common that people, when they are in healthcare, they think they need to be a doctor. But no, obviously, if you're in tech health, in the tech industry, in the healthcare ecosystem, obviously you need to know the tech aspect because you need to take care of that.
Fern Cowan:
Right. The app works and all of that. But from a clinical perspective, I think that outlier approach can actually be beneficial.
Demetrios:
Yeah, you're getting decentralized learning in a way.
Fern Cowan:
Yeah.
Demetrios:
You're getting many people giving you feedback, as opposed to just one who is opinionated because they do it this way and this is how everyone should do it.
Fern Cowan:
Exactly. And I don't know, actually just responding to your question. Yeah. The reason why I joined helcare is, again, based on the inventory and my experience and the things that I know, and that's how I've been actually increasing my knowledge. That's pretty much it.
Demetrios:
And what are some of these workflows that you're helping expedite?
Fern Cowan:
Yeah, sure. So we started with clinical workflows, meaning just nodes, meaning soap nodes, hmp nodes, all of that. But really, the way we broke through or the things that we're doing different and something that we're not doing that our computers are actually doing, or their mistake was they were trying to hard code a specific note. I want to try to use an analogy. Let's say that the tech companies and the applications were trying to make a default note for doctors, basically automate a specific note with specific parameters. Right. The problem lies that every doctor not only is a different specialty, but they also have different writing styles. So DeepCura, what it makes it truly different is that we give the entire freedom to the physician to decide how exactly they want to note what header, what clinical headers they want to add, meaning family history, social history, and another more medical jargon type of headers.
Fern Cowan:
I will say to not confuse people, but not only that, we even go deeper in the layer, meaning that on each clinical header you can add sub instructions, so you get the note and then you also instruct how you want specific headers, meaning you want it in bullet points, number list. And then now we're going to even in an extra layer that no one is doing, which is basically analytics per header. So you can say for family history, give me an analytical spreadsheet for this header and then the AI, the way it's able to make that spreadsheet is based on the patient conversation. So every time we try to go deeper and deeper, and I think that's our main edge, the only downside for this approach is that doctors, they need to have small degree of tech savviness to know how to write prompts and all of that. But that's why we have clinical prompt engineering doctors that can help them with the prompts and how to make and automate their nodes. Now we're just doing more things like telehealth and presharting and cloning the voices of the doctors if they want to automate most of that node and then they come in and then they finish it. So, yeah, pretty cool things.
Demetrios:
Oh, wow.
Demetrios:
When you say the analytical piece of it, do you mean that you want to know how many times that doctor or other doctors have looked at this person's data?
Fern Cowan:
When I say that, well, analytical aspect can entail many things. Right. But I'm going to give you an example. So, pretty popular one, Reno. Yeah, it's complex. So let's say if I go to the lab header, right, the lab results usually in the notes, it will be in bullet points. Most of our competitors, they do like in bullet points, it will summarize the lab results, right. But what you can do in DeepCura is you can add a sub instruction and say, give me an analytical spreadsheet of the lab results.
Fern Cowan:
And what it will do is it will analyze and it will create an excel spreadsheet, or however you want to call it, a table. It will create a table with all of the values. So basically, you have the healthcare data visualized in the way you prefer. So you can have bullet points or you can have tables. So basically, that is the point that we give the option and the freedom doctors to decide how they want to visualize their notes. So that's pretty much it.
Demetrios:
Wow. Okay. Healthcare seems like, and working with doctors in general seems like it's one of these areas that is difficult to break into as far as getting sales. And you did touch on it a minute ago when you said when you're in, then people start talking and then you get that word of mouth, especially if you have a product that people love. And it's inspirational to see how you have been able to focus so much on building a product that doctors love. And it's catching on because of that. And then you did not let obstacles get in your way whenever they arose. It's like there's no voice in your head, I guess, that's saying, I can't do that.
Demetrios:
You are really, in a way, you just overcome any kind of voice that is trying to deny you the ability of being able to do something which I find immensely inspirational.
Fern Cowan:
I appreciate that very much, man. Basically, the reason why I like it so much, for me, is most of my competitors, they have like an average of like $40 million in funding and all of that. So we are a very small team. And the reason when I get so many messages from doctors or clients saying that they have tried X solution, and they say that they like way more our solution, and when they say that, I don't know if it's like, I think it's just like competitiveness in a good way. I just feel very excited, like a video game, right? So I'm like, wow, we're doing this with very small resources. But you know what? I think now, in this era of AI, being small is definitely an advantage in terms of speed. I mean, if you have a small group of ten X engineers, you can definitely outperform a public company. Right now, the reason why is because AI is like every week, every month has these incremental levels of innovation.
Fern Cowan:
So basically, AI right now is redefining even the structures of everything. So what happens is when you change the entire foundation of a public company, or how you can produce a service and you have these kind of breakthroughs, you need really minimal resources to outperform even entire companies because the entire foundation has been restructured from an engineer perspective or a technological perspective. So that is the main age, basically, you really have to redefine yourself and readapt your company, your entire company, like every month, I guess, or even faster, because as I mentioned, including Deepgram, right. We're fans of Deepgram. You guys are every month releasing new features. So you really need to keep on with that and integrate. I think it's just like a game of who is the fastest on integrating all of these new solutions. And really that comes down into benefiting the end user.
Fern Cowan:
Right? And your users, once your users, they are aware that you have that speed of like, oh, wait a second, these guys this week also integrated this, or they have this new feature, they're like, oh, wow. So they trust you that you won't let them down. But if they are in another company and they see the same product, same product, then I think that might lead them to change, to use a different product. But yes, in a nutshell, that's why I wanted to say that the reason why I like it so much, this industry, and for me, this is just like a game. Obviously, healthcare is not a game. That's why from day one, addressing one of the other topics you mentioned, it's harder to enter into this industry because regulations are harder, meaning like HIPAA compliance, it's very hard to achieve. It took me like two months of studying and I'm being serious. You can just Google or ask, like, you have to make so many controls from a technological perspective on AWS, it's like a whole nightmare if you start from zero.
Fern Cowan:
So we were able to do that from the beginning with the help of really experienced people. But anyway, I just want to say that I just really like it a lot because we're a small team, but we are powerful and we have AI now. So that's a huge edge in today's world.
Demetrios:
Well, it goes back to what I was saying. With even HIPAA compliance, you wouldn't let that stand in your way. And I wonder, have you had any other trouble when it comes to regulation.
Fern Cowan:
Right now, the two major ones are always like HIPAA compliance and FDA. So also, another thing that only us and another competitor, well, it's not a direct competitor because they're really just focused on AI assisted diagnosis. Well, DeepCura was also the first one on being bold in the sense of, we're not only going to automate your notes, but also we're going to give you suggestions like AI assisted diagnosis based on the patient. So luckily the FDA has clearly defined a whole paper where they tell you the rules of how you can make an AI assist diagnosis app without problems. And the formula in a nutshell is basically first of all, it has to be non device. Your application cannot actually be hardware, so it has to be non device. And also all the advice it has to come from sources. Meaning the way we have it in DeepCura is when you make an account, we force you to add context from your preference.
Fern Cowan:
Here's another. Okay, I'm going to need like 5 seconds because it's a little complex. So whatever competitors, their hypothesis is, okay, we're going to make the clinical knowledge base, right? And we're going to force doctors trust or black box and we're going to give them the treatment plans, right. The approach that DeepCura proposes is way different and more transparent in the sense that we give you a cure, the AI model, but your responsibility is to bring your own scientific and clinical knowledge base. Because the reason why is every doctor they trust a different source. Some of them they trust up to date other dynamed. So basically everyone has a different bible, right, or different clinical religion, if you want to call like that. So it's very hard to convince to a doctor and tell them like, oh, you're going to trust our AI and it has or bias.
Fern Cowan:
It's going to be very hard. One time we tried to do something like that with popmed and doctors say like popmed, you're out of your mind. We're not going to use that. Never. Because for them, popmed is like a joke, right? They don't like that. So in a nutshell, the approach that we are proposing is you bring obviously your patient recording and you bring your knowledge base that you curate yourself. So basically deep cura uses that as a context and the patient recording, and that's how it's able to propose the AI clinical plans. And the FDA is okay with that because basically you're adding scientific sources.
Fern Cowan:
So you basically have a good foundation, like a legitimate clinical foundation. So that's how we were able to propose that. But yes, in a nutshell, FDA and HIPAA compliance are definitely the top two ones and the rest is just the doctors, they're pretty aware when they assign the terms and conditions that they have to supervise the note. We don't provide medical advice. The AI basically is just there to summarize and give them structure the data based on their prompt, their direction. So they're okay with that. So that's pretty much how everything runs.
Demetrios:
Now, you mentioned doctors. Sometimes they still need to be good at writing prompts. So does that mean that they are there and it's on them to make sure that they create the best prompts to do what they want to do? Or do you give them templates for prompts and then they can execute them? How does that work?
Fern Cowan:
Sure. So once you go to Cura, we have a section called how to mimic your writing style. And it's great because it includes templates by a specialty of some of our doctors that are really good by nature. They create some amazing prompts and the quality of their notes are like, wow, even better, it outperforms, in some cases, most AI medical scribes. That's what they have told us. But that's because the doctor has a really good understanding of how prompt engineering works. And in a nutshell, for the people that don't know what prompt engineering is, it's how good you are to give the AI instructions, right. Which in a nutshell, again, is just giving clear instructions that the AI can understand, just plain and simple.
Fern Cowan:
But obviously, when you're talking about notes with more than psychiatrists, they have notes with 100 headers or something like that, so extremely long notes. So some of our users are really good in psychiatry for that, but we have the templates and we also have, again, the doctor. We actually hire doctors on our end. And so it's like doctor meets doctor and they talk with their same language, but the difference is that our doctors are trained with this prompt engineering stuff, so they can pass and they can translate with their own lingo and they can speak the same language and everyone is happy. And that formula has proven quite good so that they understand how to make the notes in a great way and.
Demetrios:
You are recording the conversations, and then the doctor throws in some kind of a prompt to summarize the conversation transcript and then have the notes from that conversation. Is that what it is? So the doctor patient meet, you record that conversation, then there's a summary of that conversation that the doctor can have.
Fern Cowan:
Yeah. So let me tell you the workflow. Basically, we have many advantages over Chat GPT. So one of them is prompt shortcuts, which I don't even know why Chat GPT has not employed this, which is basically, if you're using.
Demetrios:
Don't say it too loud.
Fern Cowan:
Yeah, right. But no, no, we have many things. Here's another thing that I'm not really concerned about, is that HePA compliance, right? You really need heap of compliance to sell these things to the healthcare industry. But in next following weeks, we are going to release so many features that are so like healthcare specific, there's no way. Well, who knows, right? But hopefully Shajibd won't employ these features. But anyway, what we do is we have the prompt shortcuts. So basically you create all of your prompts, which they very detailed, include the instructions for all of the headers. Or if you have very simple notes, you can just make very simple prompts, but they can also include your knowledge base.
Fern Cowan:
And obviously right now we're using an engine that has like 100,000 tokens, so you can add entire articles as context along with your notes. So basically the whole point is that with a click you load your prompt and that's it. So the process is you start a recording, stop the recording, load your prompt, you have a list of prompts that you save and then you submit, and then basically you will get back the transcript, in this case accurately processed by deep grant, which is one of the best in the industry. And something that we add on top of that is we scan the transcript and we highlight the high liability parameters, meaning symptoms, meaning insurance. So all of that stuff is highlighted in different colors. And then the AI takes a look at the transcript and then makes the note based on your prompt. So that's the workflow. And then you can do more cool stuff.
Fern Cowan:
Like I said, actually something that we're doing different that Chat GPT, there's no way they're going to do this, I'm so sure is that we're not using, how can I say this? We have the thread, the same thread, but every time you make a note, it has no knowledge of previous context. But you can activate the context with a button if you want, by saying, ask your entire thread. And I'm going to tell you, the benefits for this extremely big, for the healthcare industry, which is as you know, the more you use Chat GPT in a know, it contaminates itself with past once. Let's say you are in the thread. Actually this happens to me, I'm quoting sometimes, and in the 6th time that I click submit on Chat GPT, it gives me a very lame, or like a very low quality response. So I'm like oh no, I have to restart the thread and I have to pass the context again. So obviously you got to do this in healthcare it will be a disaster. So that's why we employ this different architecture where you can manually if you want, you have that option to add the context or not.
Fern Cowan:
And the great thing that this allows you is that with the bottom you can make a whole document like a discharge summary, and you can load all of the notes from the thread. Let's say that you saw the patient last week and also two days ago and today. So you can load all of those threads as context, and then you can create new documents for insurance or for whatever purpose. But in this case, the context is being triggered manually by the physician, and this way every time they make a note, it's completely clean and the output is like virgin. It's like a funny word choice that I'm using, but it's completely virgin, so it's very hard to hallucinate in this case.
Demetrios:
I love hearing about this. I really appreciate you talking to me about it and I'm so happy for your success. I look forward to getting updates and seeing more of what you all are doing in the startup community that we've got going. And this has been awesome, man. I told you before, but I'm going to say it again. It's inspirational to talk to you and see how you don't take no for an answer and you're willing to go and overcome any challenge that you're hit with.
Fern Cowan:
Oh man, you're such an amazing host. I'm really thankful and you will do great as a host anywhere, anywhere. So I really appreciate that. And yeah, also thank you to Deepgram for all the support. You guys have been wonderful and it plays a huge role on our success. So very thankful for the quality of the technology and how that is benefiting our company and also our use research as well.
Demetrios:
We'll talk to you soon then, man.
Fern Cowan:
Sounds great. Let's keep in touch.
