AI for eSTAR: Faster FDA 510(k) Submissions

Listen.

Hello, everyone, and welcome to Friday in Focus. I'm super happy to be here and, to introduce today's speakers, who are from left to right. James. Bye.

Hey, guys.

a niche.

Hi.

And they're straight to us from Imperial College in London, and they have developed a tool for submitting 510Ks. I'm gonna let them talk about it in a second.

Make me co-host again, because I cannot, have people.

Working on it. Hang on, sorry. Okay, while I'm doing that…

We can tell you a bit, we can…

do that.

Yeah, okay, cool. So, I guess, yeah, I'm Seth, I'm in my final year, of Imperial, studying computer science. I'm Anish, I'm also in my final year studying biomedical engineering. Yeah, and I'm James, also studying, in my final year for a computer science degree. Yeah, and I guess we've been friends since first year. We've all kind of been interested in, like, starting a company together. So kind of back in November time, we kind of all got together, and we were like, it's final year, but ultimately. We don't think we want to go into this kind of graduate market post-university. So we're like, okay, let's, like.

Build a company together. And yeah, I guess on the med tech front. Yeah, so, something which I had seen working at this biotech firm called Oxford Nanopore. and other med tech startups, was that when we were submitting to the regulatory bodies, both in the EU and in the US, it could be quite a lengthy process where regulatory hurdles were kind of overlooked in both development and when it came to submission, which kind of added a lot of R&D costs and a lot of pain for both. the submitters and the med tech company themselves. So, when we kind of explored this problem space a little bit further, we then realized that it seemed to be, kind of.

Quite a larger problem than we initially thought, so then we wanted to try and build something to kind of help this problem. Yeah, and then, so basically we just kind of did loads of outreach and started speaking to loads of different people. I mean, Martin King was someone that we first spoke to, And we got, like, a lot of good information about that. Particularly, one of the things that we kind of came across was this E-Star document. We all kind of had a play around with it, and it was kind of… I mean, it, like, does the job, but it's kind of pretty rudimentary, and it's built on kind of… technology from, like, you know, 25 years ago, so it can be a bit, kind of, buggy, and, you know, there are a few different problems with it. So we just felt that we could, kind of, streamline a lot of this, kind of.

application pathway, and so we kind of decided to build a tool, which we called Regulos, which basically helps… it applies AI to help, kind of. First of all. do almost like a gap analysis of the kind of… let's say you've got to the point where you're actually creating a 510K submission, and it basically does a bit of a gap analysis to check the areas where, you need to kind of either go back to your company and produce some intermediate documents, and then finally actually, like. create this kind of lengthy document itself. And so, yeah, I guess that's a bit of a kind of run-through, and yeah, it'd be good to show you the product as well.

So what is the website? It's regulos.

Yeah, so our landing page is regulos.me. Put it in the chat, so people can kind of check it out. Host and panelists, ReadyLoss.me. Oh, yeah, nice. And then the actual app itself is… requires, kind of, login details and stuff, so, I guess if anyone's kind of interested, we can then send over some login details, and they can play around with the app itself. Yeah.

Yeah, just have them write interested in the chat, and they'll… well, at some point during the call, we'll ask them that. show… Backing up a little bit, you know… How did… how did you guys end up in biomedical engineering? You know, like, I mean… It's like, I mean, every child's dream, really, is to be a biomedical engineer. Why don't… why don't we start on the right with the niche?

Yeah, I mean, well, I was always quite interested in, actually computer science growing up, but, I became very keen on the DaVinci systems that Intuitive made for surgical robotics. And then… I thought that, in terms of, like, what I wanted to do, which is kind of combine my, like, computational skills, along with, kind of, health impact socially, I thought biomed engineering seemed like a cool space to be in, and a potentially degree to study as well. That's interesting.

Seth?

I guess from my side, so I… I've always been kind of techie, playing around with computers, and then when I was 16, I got diagnosed with type 1 diabetes, which I had no, like, idea about beforehand, and kind of how that worked, and so that was a big, kind of. learning curve, a lot of exposure to the, kind of, medical system. And so I think since then. I've been… like, I've wanted to kind of build something within that space, so… I guess when it came to, like, discussions with, these guys, it, like, made… it was a good kind of interception for us to focus on. Yeah. Yeah, I think for me, there's several things going back. I mean, I'm originally from South Africa, which is a bit of a different, like, space, technically speaking, but there's a lot of, like, you know, interesting kind of medical device applications there that I was learning from in terms of, like, more specialized things in, like, trauma medicine. I just thought were cool, but that was, like, as a child in the background.

But then, basically, I came to Imperial, and I met Seth and Anish, and I lived with Anish for a long time. We'd always talk about, kind of, cool things we were seeing. But then, particularly, I actually lost hearing in one year, and I got to gain a hearing aid, and I started messing around with that, because it's a piece of tech that, like, I'm wearing with me 24-7, like, it's really cool to play around with. So I was playing around and making software for that, and just kind of… seeing what could be done, and I just got really, really into it, and then it just kind of, like, felt like a very natural thing. It was like a way… we're all really great friends, you know, I literally lived with Anish for, like, 2 years, and it just felt like a really natural place for us all to, like, convene on to work in biomedical space.

Yep.

That's great. So, I mean, yeah, the way that… the way that we come to this is… very few people on this call came to it through a straightforward path, but, you know, you guys are coming to it right out of college, and, so that's awesome. Does anybody, in the audience have any questions for them, or should we just get straight into looking at the product?

You guys were the ones that spoke a couple of weeks ago, right?

Yeah.

That was a mess.

Nope.

You know, this is a… this is a brand new crowd. That was… that was, okay, I'm blocking.

It was another AI tool that did something with… when there were young people that had invented it.

Right, yeah, they were young, but they weren't quite as young.

Okay, well, I'm gonna ask them the same question. What do you understand about your competitive landscape? Like, are you familiar with SMVA? Because they have the exact same startup story as you do.

So I've… the name rings a bell, I'm not sure about that particular kind of company. I mean, I think, from our point of view, like. there's obviously the, kind of, historic landscape of you will have, kind of, a lot of, kind of, larger consultancies doing this. I mean, in terms of the, like, startup space, I think it is… everything is so new at the moment that we don't feel particularly put off by other companies trying to do the same thing. I think it's, like, gonna all be about execution and who makes the best tool. best tool, and then actually kind of markets it well. So, like, we have come across a few different companies, like the one you just mentioned, sure. But I think on the whole, we're kind of just going, let's just…

Make the best tool we possibly can. Yeah, and just to add up upon that, I mean…

If we were coaching a startup right now, like a real med tech startup, like, you're already hitting some big walls of somebody that's gonna be trying to mentor you or invest in you.

But not understanding.

Having a depth of that… that level of understanding. So I'm just, like, I'm just trying to, like, coach you guys through, like. you know, you have to also pretend you're on the other side, because you are, you're gonna be asking for investment soon, I would assume. And so, like, you've got to treat it like a… like a business, and… because somebody… nobody's going to give you money if you don't understand these things about the market.

Yeah, I think just to add on to what Seth was saying earlier, the competitors, a lot of them are focused on some of the other parts of the process, not necessarily the final submission and the document generation around that. And also, I think a key bit of our product that we.

I don't… see, that's where I think you don't understand the landscape, because I know several companies that do that. And it's, like I say, SMVA even specifically says that they build your E-Star for you. So, so we, we don't… I'm not trying to, like.

Welcome to Shark Tank.

Yeah.

Yeah, so I'll stop, but just, you know, bookmark those things and take them as somebody that, wants to… wants you to be successful. And knowing those things are important to being successful in this space.

I think what Michelle's trying to say in a delicate way is, like, Michelle and I have very much overlapping specialties. But because we know each other, we know the business, and we talk to each other, instead of competing, we refer to each… we refer competitors to each other. So by not understanding the landscape and your competition, you're kind of amputating opportunities that you can improve or be better than your competitors. I think that's what you're trying to say, Michelle.

It's… to a point.

No point, yeah, but I'm just like, you're… one of the things is, is… I'm in my 50s, you know, and there's other people on here, we're seasoned, we've been doing this a long time. My suggestion is, if you want to be successful, understand how you are going to be unique, and how you… how you compare to other people. That's my… that's… I think that's what I would say.

Yeah.

Okay, let's look at the tool, let's look at the tool, please.

Thank you.

Look, I think maybe some to step back and recontextualize is probably going to be useful here. So, what we're… we're going to show you the tool, but I think the main thing that we want to get out of this experience, and what we want to show to you and get from you, is we're trying to get interest in people that want to partner with us and help us build like, the correct tool for this problem. We know there's a lot of problems in this space, and we put in the legwork to, you know, make an app live, and we want to work with people in order to make, something that's specific to that problem. Again, we're entering the space, and we need a little bit of that expertise. So what we're looking for now is, like, the current tool, what it can already help you with is

if you're working on an e-Star, and you need to, like, collaborate with other people, and you need to generate a document, and you need to kind of take a lot of the manual work out, we've already got something that can help you there. But we do want to build the product and push it out, like, we're still early stages, but the main thing we want to show you here is that You know, we've got something that we can work with you, and kind of And get us to that next level.

I can't wait to see it. Cool. Thank you.

We'll, I'll share my screen and show you what we have so far. Okay… How is it? Any idea? I want to open up a new tab. Huh?

Maria E. Donawa, MD: It should be at the bottom. Okay.

Yeah, yeah, I was just making sure. Can you guys see that okay? Oh my god, okay. Can you guys see my screen?

Yes.

Okay.

Maria E. Donawa, MD: Yes.

Listen, Okay, so let's say we've… you're kind of getting towards the point where you're ready to actually create this submission. So, let's say, for the sake of argument, you're creating a cardiac sensor, and at this point, we're now being asked to upload our technical documents, so that's going to be, basically. all of the documents that are related to this medical device. So, like, risk assessments, clinical results, manufacturing documents, anything and everything in kind of all of the different formats, and you can just upload that as the kind of context for this project. So we've now created a project, and we're in this, kind of, our Easter, let's call it. And so you kind of… you'll probably recognize this structure, it's just kind of in a kind of slightly different interface. So if we click into one of these categories and click generate category.

What this is doing is now using all of those documents that you've uploaded, and it's beginning to automatically fill in, this whole document. So as you can see, it's kind of… some of these have gone green, it's kind of filled in the basic information. If we go to, let's say, device description, and click generate category again, it's going to begin to generate these things. The key thing to note here is that we're only going to generate… an answer if there is sufficient context. So, what I mean… what we mean by that is we feel confident in that answer that we're generating. And every single sentence that we do generate is cited back to the source document, so there's very limited, like, AI hallucinations, it's very much a kind of,

You, like, you will only get something if you've already, kind of. kind of produce something intermediate. So, as you can see here, we've kind of begun to… you can see some of them are green, some of them are red, and we've, like, generated something for each of these answers. If we go to, let's say, device description summary, we've begun to kind of… we've generated an initial, kind of, summary for this document. a key thing to note is we've got these validation checks, and what these validation checks do is it's kind of… we've taken information online about the kind of FDA guidance, and for each of these sections, trying to embed it. So, this one is passing, it says the text provides a device description summary,

But this one, for example, comprehensive device description, it says, provided text fails to fully satisfy the requirements, elects specific details such as labeled pictorial representations. So what we could then do is go into our AI assistant, which then has access to all of your documents, this initial answer, and say, kind of, I don't know, what, kind of pictures, am I missing? And so you can then see how this kind of becomes a, of the device, so you'd have to then upload those pictures. But you can see how this kind of becomes a review-type process, so we will try and do some of the initial work for you, but obviously the domain knowledge and, like, the kind of…

I guess the, like, context that you have as either a consultant or as the, like, medical device, producer is, like, key here, and so you can then make these manual edits and get it to a point where you feel comfortable making edits within this actual. editor. As I mentioned before, every single sentence is linked back to the source document, so you can check to ensure that actually this makes sense. And so it really just becomes a kind of review process. You kind of fill everything out so you kind of get green, get to 100%, and then finally click export, and it will export it to the E-Star format, which you can then kind of review and release to submit.

Equally, I missed a point, you could import, let's say you're working on an application, or you need to make an amendment to an application, you can import that current Easter, make edits in our application, and then send it off. So that's a bit of a whistle stop, but yeah, any, any questions?

Jose, I know you… Yeah? Go ahead.

I have a question. Are you guys pre… are you guys selling this right now, or is this pre…

Yeah, yeah, we… so we're in a kind of pilot program. We're working with a number of regulatory consultants and some larger med tech companies, and basically we're doing super low prices at the moment, and we're just trying to kind of get users on board, get as much feedback, like, real feedback as we possibly can. So, yeah, this app is live. And we can send, kind of, login details immediately, and just, yeah, get people on board. Because, like, we completely take on the point of, obviously, we're not from a regulatory background, so… we're very much in a kind of… I guess, like, understanding all the context in a kind of development phase, so we're very much just trying to get it, like, throw the net out as far as we can.

So yeah. As Sean said, I think if you want to kind of chat, put… interested in the chat with your email, and we can get in touch immediately after the call.

Nope.

Yeah, yeah.

This is Dan, Dan Barreto. What do you guys do in terms of preserving privacy and proprietary information, because there's going to be input into this application, and obviously. your clients will be interested in making sure that their proprietary information is not revealed to others. What is… what are your… processes to ensure that this data protection is achieved.

Yeah, so our main, like, data storage is in, S3, which is SOC 2 compliant. It's also HIPAA compliant as well. Yeah. So, kind of, the actual data storage is in there, and then we're using all of the best practices in terms of, encryption and data transfer and stuff, so… the application is completely secure, we can't… we don't have access to the documents either, so if you're using it as a user, we can't go and view those documents. And we don't train on any of the data you use, for example? Yep. So it's kind of… Completely secure in terms of the storage, and… from our side, we don't have access to it either. Yeah.

But obviously, you will be asked a very important question by a potential client. What you just said, they are going to say, prove it to me. Where is your validation? Of those processes, so… Have you done this?

Yeah, I mean, so the, the kind of, I guess the storage that we're using is from AWS, and they… I guess the risk, in a way, is… on Amazon, like, if there was to be a kind of data leak, that's… them, and it is, like, a kind of negligible chance of that happening. So we kind of… we have… kind of pushed off the, like, security risk in that sense. I mean, James, you can… Yeah, I think I would say, like, the risk is obviously, like, we've done all the security best practices. We haven't had… if you're asking if we've had, like, an external audit in, like, we haven't. We're still piloting this project. I think we're at the point of the project we realize that this is something that would probably be valuable, and it's something we'll try and do, if people need that to happen, or if they need extra, reassurances.

I think this is really on a case-to-case basis at this point in time. I think maybe just for… it's just worth saying, like, we have no interest in trying to steal people's information or train on them and stuff. We want to make a product that's genuinely useful and helps the industry. You know, we want to see good products come out faster. So yeah, we're doing all the best things we can there. We don't have any external influences or, like. That kind of thing. But, on a case-to-case basis, if you have more concerns about that, we can.

And I'll finish by saying, if it's not documented, it's a rumor, it's a phrase that you guys want to become fully aware of. Because that's what they're gonna say, show me the documentation. So, I hear what you say. But what you say, it's more hearsay, it's a rumor, until it's really documented and proven. So, just for you to keep in mind.

Yeah, I'd like to double down on what Dan is saying, and Daniel is saying, and just plan for the audits. There is no way in hell you're going to sell something like this into a company without legal… looking at your data models, I mean, your paperwork and IT, looking at your data models to make sure the data is secure, so just expect it.

Yeah, yeah, no, that makes sense.

I have another question here. Sorry, guys. Are you, having, dedicated instances of your AI models here, or you are using just, let's say. public API. I mean, you know, ChatGPT, GLOD… Without breaching any NDA, of course.

Yeah, I mean, the reality is that we can… it depends on the requirements of the user. Like, we're kind of using a bit of, like, a… on the architecture side, we can suit that to what the user needs. We can use either public API, and then also public API with guarantee of no retraining and, like, other sort of, like, different enterprise guarantees that we can get. But all of those would, I guess, be… We could also self-host models. It depends on, I guess, the level of confidentiality. That the client would require. Yep. content.

Platform itself is checking only the, documentation, and… Generate them, or do you have, kind of, Verifying process of the company, or… Verifying the process of, you know, kind of monitoring control on the… document that you actually generate. I mean, it's a… Can you demonstrate something like this? I mean, let's go to cybersecurity, for example. I mean, what kind of.

Yeah, if we generate… So, sorry, to clarify your question, are you saying, kind of, is it a generalized framework, or are you, kind of, are we using…

Question is, is it focusing only on the documentation, or you're doing some… Verification with your customer… target customers, also with… The process itself, you know? The real-world process of how they Aligned with, you know, with this documentation of what they are doing with this documentation after they are My question, if I will formalize it and simplify it, is it just a document generator? Tool, or much more than this?

So, I mean, we have this kind of… I guess, like, AI validator, which tries to kind of, as you said, like, use the FDA context. I think a few that we got early on was trying to produce this whole document all in one, and then be like, this is perfect, is not the way to go about things. So we're very much, like, it's a kind of… human is at the forefront of this process. So, I guess, to answer your question, it's a document generator, and then the human is that kind of final port of call of… does this make sense? I need to make these here and there. But we have embedded, like, structural constraints and, like, FDA checks that we've tried to implement so that the content and the validation is something which is, hopefully, means that the user can have as little rewriting needed as possible.

Hi, can I jump in, here? Yeah. you're… the… the… I mean, like, it's brilliant, what you're putting together, but, For a medical device company. Okay, so you've created this, and then you submit, and then the FDA makes a remark, and then you've got to make those changes. For a medical device company, this… the putting together this package, yeah, you've saved a little bit of time. But, you've got to have traceability of your source documents, and that is where the… a lot of… compliance issues occur. So, there's that problem. Rather than putting in compiled documents, a PDF is a compiled document, it's a printed document.

You then have to be able to trace what happened between the first version you submitted and the second version you submitted, and that is where your content management system comes into place. And so, what you should be looking at, really, is connecting to those systems, and then this is compiling from source, rather than compiling a generated Document, because in between the time that somebody sent me the PDF to put into this, the compliance person puts, you know, all of this inside. Somebody's generated a new PDF. How do I know I've put the latest version in here? So, you're working from an output as opposed to working from source. Another question I've got from you, for you is, what LLM are you using? And then, when you start

thinking about, like, okay, well, what LLM am I using? How are you… how have you, audited that LLM? How are you sure that in… if I click that generate button 100 times, I am going to get the exact same reply 100 times. I've… I've built something similar to this, and you generate 3 times, and, you know, it changes slightly. One single comma change. is a non-compliance, you're gonna get yourself into trouble, with, with, with, submissions, so that's, like, you know, I think some of the questions that'll come out, like, you know, that you've already been asked, about privacy, you've got to be able to prove

prove, you know, your LLM audit. There's a whole load of things, especially… you guys are in the UK, right? So, you know, there's… you've stepped out… we've stepped out of, the EU, but the EU AI Act, like, you know, brings in a whole load of things that you guys are gonna have to consider, because you are acting on information from a medical device. Where do you sit on those risk levels? How have you audited, and how can you prove everything to do with the L&M and the generation, and the AI part of it, in… with regard to the EU AI Act, and any other regulations that are going to start coming up. And, I mean, like, that's the first published one.

But ev- you know, different regions are going to start coming out with their different, you know, regulations. You're going to have to be reactive to that. You're not only going to be submitting to one, to… okay, this is East Star, so you're submitting to, like, you know, the US. Companies, medical device companies, are selling worldwide, and not just submitting to the FDA.

Yep. So I guess, to answer your first two parts, the first question to do with, I guess, like, going back to the source documents and kind of how we ordered that, I mean, I think… Currently, our, like, context management system is relatively simple. I think, like, longer term, we want to be looking at, kind of. probably, like, working with a, kind of, these QMS systems, as you said, to kind of plug into them, so it becomes a case of almost, like, you have your QMS system, you have all of the documents managed, and then it is just a case of you click generate, and you get 95%, and you can read over the whole thing. We thought, in order to get a product

straight to market, we were like, we'll go with this approach of, kind of, you're assuming a level of… you have some of these documents, kind of, ready to go. In terms of your second question on what model we… this exact version's using Gemini, As James is saying, it's.

UI.

Pardon?

I recognize the UI.

Yeah, yeah, so, and, like, it's, I think, in terms of… it can be a case-to-case basis of we can kind of use any model we want, so if a user has a preference, that's not a big problem. In terms of the, kind of, validating… getting the same response 100 times over? I mean, James, you might… I mean, obviously, this is something we're working on, right? So, in the current pilot, the current controls for that are, like, essentially a bunch of rails that the AI must stay on, and also there's a notion of, like, in terms of change management, we do have, like. we didn't really show it, but there is, like, a change random system across, like, all of the versions of the document. But also, when you generate something new with the AI, like, let's say you added another source document, and then you asked it to regenerate, like, the new thing that gets changed would be shown side by side, so it's more like, we understand that

every kind of solution to this would have to be, like, a solution in depth. Like, there are some rails happening, but also, we want to make it as easy as possible for, someone reviewing to then, you know, be confident in everything that's being generated.

Yeah, sorry, just on that point, you… that's the wrong, I'm saying, with the exact same documents. If you click generate 100 times, will you get the same answer 100 times? You're gonna have to prove that. You're gonna have to have that… you're gonna… you know, have to get that audited. And with the exact same documents, I can bet you that if you click generate again 5 times, Gemini is gonna, at some point, change, for the same set of documents, with the same guardrails, it's gonna change something somewhere. And give it a go. And also, like, you know, it's quite easy. Add, like, you know, a couple of other models in there as well, so that you can start, you know,

testing different models, because once you've built this out and you've focused on Gemini, as soon as you plug OpenAI into it, the answers are going to be different. Those percentages calculated are gonna come out different.

But let's say… sorry, for the question, but let's say they are solving this, okay? they are going to run their own, even small models, okay? And based on my experience, once you are running a dedicated small… model, not a huge general-purpose LLM, you can achieve these kind of requests and answers. Should not be a problem. So, technically, I think all can be solved. I mean, I don't know you guys, but.

Hmm.

my experience, so… the question, at the end of the day, who is your audience here? I mean… Have you done some, you know, product-market feed?

Well, that's why they're here. They're really here to, you know, sort of debut the product, to get feedback. I would encourage you guys to keep the feedback a little bit more positive. The, I'm not saying that you shouldn't ask questions, however, I think that what we have the opportunity to see here is, you know, some guys that are, you know, straight out of university, who are building a first-time product, and I think that we want to, encourage them to move forward, rather than throw up every obstacle. The…

I think Sean, though.

They see every obstacle before they get set on, like, you know, one…

Will you…

thing, you know? I'm giving… my feedback is for things that you should go back and look at, you know?

Because the people you have on the call are pretty seasoned people, honestly, and we've seen all of these problems happening, and we're just, I think, simply trying to make sure you don't stumble on the same things as other people have stumbled, we've all had problems with. We're… I mean, and I'm… again, I'm more positive about stepping forward and making sure that you don't have those same problems is really what we're trying to do, Sean. I don't think we're trying to be negative about it. I think we're just trying to… Make sure you don't have those same issues.

I think, yeah, I think from our side, like, we're, like, we're very open about the fact that, like, we don't have all of the, kind of. like, regulatory understanding and knowledge. So I think from our side, we were like, okay, let's get a product out. We know there are some holes in it, and there are things that we are going to need to focus on. And so, like, this is, like, really helpful, because we kind of have this, like, more of this information now. We can then go and make the relevant changes. And come back to you with those answers. And I think, also, kind of. For anyone who is interested in, kind of, working with us, like.

We will give you all of our private emails, and we'll be able to respond to you immediately, and if you have a specific kind of request. we'll implement that as quickly as we can, and kind of, like, we work on a very quick basis, so I think, to be honest, like, the constructive feedback is, like, really helpful, and would, like. we want to be practical of this. We're not just gonna… gonna put our heads down and work under incorrect assumptions. I feel like that's the…

The rest of us, loud most, have talked, maybe Maria has something to say.

Maria E. Donawa, MD: Okay, thank you. Can you… you can hear me?

It is.

Maria E. Donawa, MD: Good. Maria E. Donawa, MD: My question is… Maria E. Donawa, MD: your target, purchaser, or the companies that would be interested in this, may not be very sophisticated software people that are making comments on this. So, my question is. Maria E. Donawa, MD: Are you comfortable with describing to a user that doesn't have a lot of sophisticated AI background. Maria E. Donawa, MD: the advantage of this, of using this versus just sitting down with the ER system and struggling, or working, or maybe even doing okay with the E-Star system. So. Maria E. Donawa, MD: I didn't hear at the beginning, maybe I missed it, the advantage of using this tool

Maria E. Donawa, MD: Versus just not using the tool, because… Maria E. Donawa, MD: you may be… I think you're gonna be talking to clients that… Maria E. Donawa, MD: don't have any experience, maybe even with another one, maybe they haven't shopped around yet, and eventually they will, but they need to understand the difference between working without this and working with this. And then after you make a comment, I just have a… I have a comment after that.

Yeah, I mean, I think, like, yeah, if you were to use the e-star and there's the workflow of, kind of. going through each of these questions, going to the internet, understanding the context. If we're working with someone who has very limited regulatory experience, let's say, kind of going to the internet, understanding the context, also, like, going through these kind of tens to hundreds of, like, documents related to their device. Like, that in itself is a kind of… very, like… time-consuming task. So I think the key idea that we kind of want to help with is first of all, if you try and generate an output, it will direct you to the key documents for that particular section that you need to rely on and look at. And it kind of circumvents just this kind of super manual process of look at a document.

searching the internet what this question's actually asking for, trying to understand all of this different context and bring it together. We have this information readily available, so you can kind of very, like, easily just go through each question and… just speed up the process, like, we do think you could kind of… Half the time that it takes to produce these documents, this, like, application with our, with our, kind of, platform.

Maria E. Donawa, MD: So, so basically, you're concentrating on the saving of time.

Yeah. Are you…

Maria E. Donawa, MD: Are you also concentrating on increased accuracy, or any aspects of the increased quality of the submission, or…

Yeah, I mean, I think we can take a step back here and be like, okay, well, really the, the, the… the product has a couple key components. There's the fact that this is, like, a web app and, like, hosted, and people, like, you have a login, and it's all saved online, and you can collaborate, everyone else, someone else can, like, log into the account, you can collaborate on the same thing. These are, like, excluding any of the AI features, these are just, like. things that you wouldn't get with an eStore that you'd be getting by using this platform, which is a bit more of a modern, like, feel and interface, and, you know, a better way to store things, instead of having, like, a really long Adobe Pro document open that you have to, like, manually add all the attachments to and stuff. Being able to see all that information at once, that's just, like.

a baseline thing that we're providing. And then in terms of on top of that, with the AI, there's the AI to generate, which we think saves time, it stops people from having to copy-paste. There's the AI even just from telling you, like, what documents you should be looking at. It's almost like it's just giving some bit of help. That's maybe not, That would be more useful if you were newer to running these kind of documents, and you would have, like, a bit more of a heads-up. of just, like, for this document, you know, for this part of the document, the FDA is recommending that the submission should look like this, and something you wouldn't necessarily know, because I think it's just a really intimidating space to enter. So I think that there's some value provided there. But then the layout on top of that, and, like, here's what we're trying to push, is, you know, we want AI to be able to generate these, and then validate them, as much as possible, and check for extra holes.

But maybe, like, the human… could do, or maybe they don't want to do, but if they… that, in theory, could do, but don't have the time to. We think it can be a very time-consuming process, and we want to try and give, like, extra levels of reassurance. But these are all, like, there's, like, levels to this, and this is obviously a work-in-progress product, but that's, like, where we're trying to add value at each level.

Okay.

Maria E. Donawa, MD: No, I'm sorry, I have a second part to this. Sure. I'll be quick. Maria E. Donawa, MD: So, that communication with the potential user is critical. Now, I think the points initially made about the fact that you need to understand what the competitive landscape is, is a really good one, but there's a competitive landscape to everything. Maria E. Donawa, MD: So, understanding, basically, what's out there, and if another program does this the same, or better, or whatever, you have to know that, because otherwise… but I… to me, that's… that's a part of it, but it's not the end-all, be-all, because everyone…

Maria E. Donawa, MD: has their own aspect that they're concentrating on. You could have two of these tools, and they're… Maria E. Donawa, MD: almost the same, you… it will be your responsibility to, understand what your tool… the advantage that your tool has. The fact that they're out there for me is not a big deal. They're gonna be a lot of these, of everything, every tool. They're gonna be multiples. And you're gonna have to compete with that. Maria E. Donawa, MD: So, if you don't right now know that, I do think it's important for you to eventually know that, but… Maria E. Donawa, MD: I'm not… I mean, that's the life that we're living now.

Maria E. Donawa, MD: That's it, I'm done.

Thank you, Maria.

Hi, Maria, haven't seen you in a while. Okay, so… so I have kind of an on-the-ground question. So, this is… this great looks like you pull a lot of information, put together. What does it look like in the E-Star? Because, you know, we submit the E-Star file. not… It's not submitted here, so… what does it look like in the E-Star, and how do you deal with the fact that every time you answer a question, something else pops up in the E-Star, and sometimes nobody even knows, has never seen it before, because devices are very different. So, how do you deal with the hidden pieces of E-Star?

Yeah, so in terms of the hidden stuff, it's probably easier if I create a project. That's all… it has the exact same functionality as the Easter. So, if we create a fresh project… project… And we look at, let's say, administrative information, currently it shows nothing. If I then fill in the US FDA English, that should then pop up with some of these questions. So, that functionality is the exact same as the E-Star. In terms of exporting to the E-Star. every single text, answer, multi-select option is automatically populated in the eStar. You can send through a kind of video of that, like, a kind of proof of that after, if that helps. But yeah, all of those are then filled in, and then any kind of attachments that you need to.

attached to the E-Star, which we've generated for you. you'll get a folder of them, and then you can manually go through and kind of add each of those attachments. But we validated that all of the, kind of, import and export works as it should.

Okay, so… so the… see if I heard you right. So… once the auto… these, I guess, the generated aspects of… that go into the E-Star. But then, if I'm using this, I have to go back in and manually attach the documents in those categories? Is that… did I hear that right?

Yeah, so, I mean, I think this is, a little bit of a side point. This is just, like, a, just, like, where we're at in the development right now. In terms of, like, what it can do, like, literally right now, it can… like, every single field that's in the normal E-Star, and every possibility, like, can be accessed in this platform, and then when you export it, it'll all fill in. I'm not sure if you've noticed in the E-Star, like, when you're trying to move it between the versions, and you click the import button, you know how it can't input, import attachments right now? That's like an E-Star limitation.

We haven't worked around that yet, we're trying to work around that. Like, it's obviously… we want it in the future. use a single button. It's just, like, as a teething problem, basically. So… For the time being, what that looks like is imports, and then you have to attach. It's something we're looking to make better. We know that that's not the most economic way to do things. Because, I mean, I think there is… ultimately, we want it to be a case of you don't actually need to use Adobe Pro at all. You could just get rid of your, like, subscription, and you have the confidence. You need to click.

Ready to go.

So, just from a time-saving standpoint. That's a… actually physically putting those in the E-Star.

I don't think the FDA is ever gonna let us do that awful, god-awful, in my opinion. form. I mean, that form is awful. And every time they update it, it takes longer and longer for the little stupid clock and wheel to…

Drop down another lovely…

Dropbox or radio button that we have to… for anybody who's done that new 6.2 form that they just came out with, it's lovely that we now have to do with the QMSR. Have you gotten to install those new lovely? SOP, if we get a drop in there now? In my opinion. There we go, I'm done.

So, there were a number of questions.

There's no way the FDA is going to let you integrate anybody's AI form, whoever's landscape is out there. There's no way anybody's gonna be able to break whatever code. they have…

or whatever else the FDA is going to break in for you to integrate anybody who's.

Professional landscapes out there.

Question, have… have you… Already proved or checked. how much time it's actually, saving? I mean, for a new startup. It can be a really, Great starting point, but the question is. How will you take it to this next level?

Yeah, so we launched two weeks ago, and we primarily focused on consultants to begin with, who are kind of using it. We haven't got… we kind of, in the next kind of week or so, like, wanting to kind of get feedback and get those numbers, but we don't have exact kind of stats of it saves 40% or 50%. And yeah, we assume that the… people with less knowledge, it should save more time. The more experience you have, and if you're creating, kind of, 5-10Ks on a day-to-day, less time, but obviously, you know, you create more of them, so…

My two cents here is just, if you can… Be in that situation that you can say, it saves this amount of time that equals this amount of money. And this will really help you.

100%, and I think also that will help with the kind of pricing element. At the moment, as we said, we're Charging a number of sound.

So let me… let me ask you guys a question. Do you… do you guys have, like, an advisory group that is basically supporting… You, as a company, meaning… a technical advisor, somebody who worked with… as an EU regulator, or as an FDA, you know, regulator. Somebody who knows about marketing and business strategy. Are you… are you just doing this on your own, or are you actually backed up by… A group of, let's say, senior people who… Who, like somebody said before, been there, done that, and could help you to sort of bring the company to the next level.

Yeah, so in terms of… from a regulatory side, we've got, kind of, a number of strong informal relations, who we're, kind of, leveraging and building very closely with. I guess, in terms of everything else, someone mentioned it before in terms of investment. We're currently speaking to a, kind of, investor who, are in America, and we've got the final call. in a week's time, if that goes well, then we'll get funding, and basically go to the US in June, and basically go full-time on this. And that will then provide a huge amount of support in terms of Actually, kind of, building a business and everything that comes with that, and obviously the actual investment will allow us to hire external people and things like that.

I think your answer was no, exactly, but… which is fine, but my suggestion to you is to sort of seek that… that type of support. I mean, it's… it's like, It's gonna be the oxygen that you're gonna need when you need to climb a step, you know? That's it.

Yeah, that's helpful, thank you. We could… should we answer some of the questions in the chat? I mean, where there are quite a few? Yeah.

I've got just a couple of things to say. So, first, congratulations, this is great. As someone who's worked in the last, 20 years or so in startups, mainly, and had my own. It's a… it's a difficult space, but it's very rewarding, so, good, for you. I do want… To just mention, really, though, is that, medical's device is very risk-averse. And normal stuff like agile methods, where we'll… we'll do things… Sort of on the go and solve problems as they come up.

Generally, just not, going to get through. You know, if you look at any medical device risk management, it is super, super thorough. Like, you know, you'll have hundreds and hundreds of lines of, these are all the potential things that can go wrong. And at one point, you… there was a statement of, it's 100% secure. I would never use that word anywhere with anything to do with the… connected to the internet. You're… you're not 100% secure. You're going to want to show You know, penetration testing and all of that stuff to show that your data is secure to a specific level.

You know, so… just be warned. I think what you've heard here is a lot of people who Work in this very risk-averse environment, and don't… I want you to, you know, to do well, but there is a lot of things that you're… isn't like a website or anything like that, where, you know what, if it goes down. Don't worry, we'll… In two weeks, we'll have a new version up, or something. It's not the same thing at all. And I think that's sort of what I would, get out of this if I was you guys. So, good luck!

Yeah, I'd like to thank everybody for their constructive advice. I want to give, Jose a minute to show what's been going on with MLVX, but before we do that, I'd like to give, James and Seth and Anish a moment to, let us know how we can help, you know, and how we can connect. Feel free to put your email addresses in there, or I can send a follow-up to anybody who would like it. Anybody who would like to speak with them privately, please put, you know. a message to me, and I'll make sure that they get it from the chat. And Yeah, I… I think you guys are doing something really neat here, and, you know, it's certainly a challenging road, and many of us have been down it.

Going to market is not easy, and there are many technical challenges that have to be surmounted. Thanks, everyone who contributed. Guys, any closing remarks?

I guess, just, we really appreciate your time today. We don't, like… we are aware we don't have the expertise in this space, and so, yeah, anyone who's kind of interested in getting in touch, we chucked our, landing page in there. Yeah, and we have a waitlist on there, so feel free to, if you're interested, to, drop us an email and connect with us. Yeah, I mean, at the end of the day, yeah, I think… It's been really great, really constructive, and it's… the amount of, like, reply and enthusiasm has been really great. I think if you're interested in trying out the tool, or working with us, telling us anything, please send us a message, and yeah, we're really receptive for anything that you can tell us. Yeah, and anyone who has put their email in will follow up immediately after this call.

So, thank you, and thank you, Sean, for hosting. Yeah, thank you. Yeah, thank you.

Now, I want to give Jose a minute for MLVX, and show you what he's been doing, because it's crazy.

Alright, hi everybody, if you haven't been here the last two weeks, as you know, Sean and I and our team launched this MLVX platform, which allows you to, like, see who the experts are in the field. If you don't have an expert profile, you know, let us know, or go on and… register, then you'll… you should be able to create one. So you can see awesome people like Sean and myself, and Michelle, and Marina, and all these people. And now you get to see some of their posts on LinkedIn, and if you have an account here, you can… you can essentially share your posts here, so that people can kind of, like, not just see whatever LinkedIn wants to serve you, but see really what the experts in the field are doing.

And then there's an AI chat, system, and kind of a cool feature that we added this week, just wanted to show you guys, is, well, here, I'll just show you. Let's see… So I'm gonna ask, what are the main EKG FDA codes? So this is an AI chat, but what's cool about it is that it pulls information from content from the experts, right? So all of the experts that I showed you, if they put in their blog posts, their webinar transcripts, their… you know, like, videos that they've put together, it'll pull information from them. The latest feature that we added this week is we're linking it to the FDA database, so now you can ask it things like, hey, what are the main EKG FDA codes? And it'll tell you they're DPS, MWJ, MLO, QDA,

Etc, and if you then said, okay, I'm developing… an EKG device that connects to the cloud. What cyber security documents do I need to prepare? Then, but everything works right, it'll start pulling information. not only from guidance documents, but also from, like, let's say, blog posts that experts have uploaded, like, you know, I write a lot about cybersecurity, so you would see that there. So it's kind of cool, like, not only is it pulling from the guidance documents from FDA, from expert content, but then now it's also linked to the FDA database, so, so you get that benefit as well, kind of bringing it all together.

We're still building this out. If you guys have features that you think, hey, this would be really helpful, like, our goal is for you to do a lot of your regulatory, kind of. AI chat stuff just in here, and it gives you an opportunity to see, you know, still typing away, because it's getting a very comprehensive response. So yeah, if you have features that you think would be helpful for you, then just let us know in the app itself. So look here, it even tells you an expert, and of course it's me, because I read a lot about cybersecurity. But if it was about another topic, it would be some of the other experts. We have over 130 experts now on here, and then the cool thing for the experts is that now, you know, I see, hey, look at this guy, what does he write about, what does he do?

And then you can see, you know, I've got more than 25 content pieces on here. So, as an expert, it's to your benefit to load content here, so that as users come in here and, you know, research topics, you're the person that gets linked to those topics. So there's Paul. Yeah, any questions?

That's amazing, Jose. Fantastic.

Awesome. Thank you. Thank you, thank you, thank you. Have fun with it. Yeah, in the app, there is a thing that says set feedback. You can send, you know, general feedback, bugs, feature requests. We're looking at this constantly. Most of the features that we're building is because somebody sent us a feature request on here, or Carrie broke something. If Carrie broke something, that's a feature.

And then… That's happened a lot.

Or if somebody asks, yeah, And then, you know, my final word, just, you know, Seth James, is it Anishh? You know, really, just to encourage you guys, I think those of us who've built startups or have been in the space, you know, like. I think this is great for you. Like, you're getting it from, like, critics who are… you want to be your customers. And so just take all their information, and take all their criticism, and don't give up, and keep building, and I think E-Star forums are terrible, and they suck, just like Carrie said earlier, so it is a pain point, and it's just a matter of getting the solution right, and then the story, and the security, and the…

You know, like everyone else has said, like, it is a very risk-averse space, so just learning how to address all those questions, you'll be in good shape.

Yeah, there you go.

That's on it.

Great. Yeah, great, thank you.

All right. Someone asked, does your data corpus include 403s and warning letters? Jose, not yet, right?

It does not. It does not at the moment. Is that… I mean, if we think that's helpful to incorporate, then… we can start pulling those in as well. And by the way, when you're chatting, if you chat and you find that it's like, hey, this really should have told me about this reference, you can ask it, like, hey, do you have this reference in your database? And if it tells you no, you can drop in a URL or a PDF, And then it'll… it'll send it to us for review, and if it's relevant, we'll accept it, and then it'll get ingested into the database. So, you know, the next time you ask a question, or anybody asks a question. So that is public, and it does warn you. It says, hey, this is going into the public database. Your actual chats, don't get pulled into the database.

But if you wanted to pull something into the database, you can do it right from the chat.

And it talks to you and argues with you, and it's kind of fun. Carrie.

Is that the carry feature?

Jose and Carrie.

You can put it in carry mode, and then it yells at you.

I just put there, what did Carrie break now? That must be a regular hashtag.

Just have me break it, I'll break it for you. I'm your worst nightmare when it comes to software, it's fun.

stuff.

Man, but Jose, this is huge, brother. Fantastic.

Yeah, glad you guys like.

I'll play.

Play around with it, give us your feedback, spread the word. We do have, like… so, by the way, also, if you want to have on it, and you want a promo code, we're doing promo codes for, like, three months. So there's a free version, you get, like, a few questions you can ask. There's a paid version, there's kind of, like, a high-level paid version, and then there's an expert type of version. Like, not versions, but, like, accounts, or profiles. They have different costs, but if you just let us know and you're willing to spread the word, we'll give you a 3-month promo code, so you'll get it for free for 3 months.

I mean, are they?

Never let you know? Just curious.

Just, just shoot us a message.

Okay. Like LinkedIn or something, or…

LinkedIn, right on here.

I'll put it right here in the chat.

Pretty much. I'll pick it up. I'll send anybody their promo code.

If you're on this call, you don't have a profile, say you want a profile. If you want a promo code so that you can upgrade, put promo code. I'll upgrade you. So, once you're upgraded, you can start uploading content to your profile, you can upload PDFs, you can upload slides, you can upload, transcripts, documents, you can link to blog articles that you've, that you've published. So, you know, Maria, your articles. Link to them. They'll be uploaded.

Maria E. Donawa, MD: Actually, I was gonna write a chat saying that I needed some help, with it. I just visited it, and I think… and I noticed that I hadn't really done anything, so I definitely need some help.

Nope. I'm the customer success team, you're looking at it. So… Well, it's 1201, Eastern Time, and I want to thank everybody for your time and attention today. I'd especially like to thank our speakers. James, Seth, and Anish for, braving the Shark Tank and, coming out here and sharing with them. I'd like to thank everybody for their very, very helpful, suggestions, and, and advice to these young entrepreneurs. I, I'm excited to see what they do, with this, and,

Wish you best of luck with funding. Round.

Thank you very much. Cheers, Sean. Thanks so much. Thanks, everyone, Al. Thank you for the advice.

Cheers.