Oura Ring’s Dr. Tanvi Jayaraman on serving women in the AI era with its first female-focused LLM, chatbot

[MUSIC] If you're running Creator and affiliate programs across separate platforms for every channel you sell on, there's a better way. Levanta helps brands and agencies run one unified program across Shopify, Amazon, and Walmart from discovery to pay out in a single platform. Find the right creators fast, activate partnerships in clicks, and see exactly what's driving revenue across every channel. And right now, when you book a demo, you'll get a $100 gift card of your choice. Stop paying for three tools to do one job. Head to levanta.io/glossy to book your demo and claim your gift card. [MUSIC] [MUSIC] Hello and welcome to the Glossy Beauty Podcast. I'm Lexi Lebsack, and today I'm joined by my colleague, Sarah Spurkfiner. Hi Sarah, how are you? I'm good, how are you? I'm doing well. I'm really excited for this episode. I spoke to someone from Aura. You have an Aura ring as well, right? I do. Yeah, I've worn it for a long time now. Oh, what is your favorite part about it? I don't know. I mean, I feel like I have a complicated relations with it. I wouldn't say I'm one of those people who feels like I have improved much because of it. Like, I sort of just exist alongside it. And I would like to improve alongside it. And for a period when it broke once, I realized that I missed it. And I had gotten so acclimated to knowing my sleep score and all these things that I feel like have become a part of so many people. Yes, we live in a bit of a bubble. But in our bubble is vernacular. Like, I was with another beauty editor the other day. And we literally were like, I slept terribly. And we both knew our sleep scores. Like, it's very much like a language that we speak. Yeah, Farah, Mia Ma has been to that with our sleep scores nowadays. We wake up and kind of share sleep scores. I just started wearing one at the end of 2025. And sleep has been the first thing that I've kind of gotten into tracking. And it's been nice. I've learned a little bit about what I require to actually get good sleep. Like, the first thing is that I really have to be in bed for like 10 hours in order to get eight, eight and a half hours of sleep. You can't just pop in and then expect to sleep the entire time, which has been really informative for me. But yeah, I have an exciting guest from Aura. And it's sort of, yeah, it's kind of exciting, but I'll just sort of get into it. Basically, I have a guest named Dr. Tanvi J. Aramun. And she is an MD and she's also the clinical lead of health AI at Aura. So we're talking about the Aura advisor today. Do you use the advisor at all? It's sort of the chatbot where you can. I have started using it and I will, I can tell you exactly how. Sure, yeah, how are you using the advisor? Because that's really what we're talking about today is the advisor. Which is for people who are unfamiliar, it's sort of like a little wellness chatbot where you can ask questions and also ask it to like, like, analyze some of your data that you're collecting. I'm very excited about this because it sounds like you're, you know, somewhat more of like, even though I've had the ring for literally years, like I feel like you're already a more informed user of it than me, but I'm trying. It's a, it's a bad sort of background goal. But I have used the advisor sort of, you know how Aura like has really impressive accuracy with like when you're getting sick or whatever. Like it says like, you have minor signs or whatever. I will be like, why do I have minor signs? Because I like really want to understand like what it's sensing. Because sometimes it'll be like, I know I'm sick and I'm like, that's so crazy. How do you know I'm sick too? Or other times it really is sensing it before I even know that something is wrong. And I want to understand what's wrong. So that's been like the way that I've used it a couple times. But I know there are like a million more ways that I could be. And I'm excited to hear this in a review. Yeah, no, I think most people are using it similarly to you. You know, you see some data or something's going on in your body or you see some sort of alert from the ring. And then you go into the advisor and kind of ask like what's going on. And this is sort of the frontier for wellness in my opinion, especially around wearables and trackables. Because now that we have all of this data, we now need an AI chat bot, I mean, or something, to basically analyze it for us and tell us what's going on. Because I can look at my aura results and I mean, it's just like Latin to me, right? It's not something that I necessarily understand. So aura kind of is a first mover in this space. And they released an update to their advisor. And it's an LLM, so a large language model, which of course is sort of the guts of all of these AI chat bots. And it's they're the first ones to type trained their LLM only on women's health. And so basically, Dr. J. Aramun, she and her team have hand selected all of the information about women's health that the LLM has been trained on. So it basically just creates like a much more accurate way to analyze your result and to learn information about wellness and health. And it's an interesting role that Dr. J. Aramun has because she is an MD from Stanford University. She also worked on AI strategy projects at Bain & Company for sort of global diagnostics for pharmaceutical companies. She also worked on Apple's clinical team, shaping like early concepts for their next gen digital health tools. And then she joined Oral last year and her whole mission there is to kind of bridge this gap between medicine, artificial intelligence, and product strategy. In today's episode, we discussed how aura sort of pulled this off. It was a big endeavor, a lot of work, a big team. We also talked about how it directly impacts aura ring users. Of course, there are millions of them. But I think that most importantly, we talked about how first movers like aura are really driving changes in wellness. And those changes are really reflective of women and women's health needs and outcomes, which of course, you and I, and most people nowadays know that women weren't really considered in a lot of the health studies and various things that happened above my pay grade. But yeah, so it's a nice sort of step to better consider women and to be a sort of leader in this space. I love that. All right, well, here's my conversation with Dr. Tanvi J. Aramun, MD, clinical lead of Health AI at aura. Dr. Tanvi J. Aramun, thank you so much for coming on to the Glossy Beauty podcast today. Thanks so much for having me. I'm super excited to chat with you, Lexi. Well, I am a recent aura ring wearer a few months ago. I started wearing one and I recently opted in on the new aura advisor custom proprietary LLM, which is really what we're talking about today. This sort of new big push from aura to better support women, which of course we know has been the backburner for many, many decades amongst the medical community. But talk to me a little bit about sort of the why here. Why, what was the problem that we're solving? Why make a sort of proprietary LLM? And what is it? Yeah, great, great question. I'm so excited to share more with you and the listeners today. I got into the idea of becoming a physician because I, the foundation wanted to help people. And I wanted to help people in a way where I could be a trusted coach for them in the room as they were grappling with a lot of really murky and ambiguous symptoms and pathologies and looking through a lot of different diagnoses and how do you parse through a lot of really complex information. You really have to find that coach in the room to help you think through it all and ask the right questions. And a huge focus of mind during medical school was actually in women's health. I actually led the women's health clinic and our free clinic when over at Stanford for a year or so. I think what we see is that women's health has long been underserved and there's kind of this one size fits all data model that we see. And so we really wanted to change that with our first custom LLM and we designed it and we've vetted all of the training data as with the clinical team with PhDs and on staff with our currently clinically practicing OBGYN. And we really wanted to make sure that everything that this LLM was trained on was actually used in the real world by physicians today. And so it's data and knowledge that's uniquely vetted by our in-house team. And a big part of building this LLM was making sure that that data was actually usable by clinicians down the road. So we really want to make sure that women are getting the guidance and the support that they need. That is truly trustworthy. And so our process really took that into account. With this model, just to give you a little bit of explanation, we really feel like we're pushing the boundaries of what's possible with AI, with wearable health data coverage. And we hope that this is the blueprint. for future potential use cases for other health conditions. I see the power of AI, why I get really excited about it as being a way to connect the dots between kind of what's the most reliable validated science with what's actually happening in your body. And we know that in women's bodies, it's kind of an end of one. You don't see the exact same experience over and over and over again. And we really hope that with this model, that we can give every single female member a unique understanding of their body and their data and connect it to clinically valid sources. To give you a little bit more of an overview, what I actually find is another really unique aspect of this launch is we're launching an aura labs. We're not launching directly to production. We think that, and this is super intentional, we think that launching in aura labs means that the women and the members that are testing this model, they get to actively provide feedback to us. They get to send us questions for consideration with how the model is performing. And the features in aura labs are considered discovery phase. They, you know, eventually if they aren't performing well and if the feedback is not great, they can be removed. And eventually they might get promoted to a permanent place in the aura app. But we really want to make sure that we're fostering a sense of collaboration between our members and between the science team and the product team and the clinical team. And I think that feedback loop is going to be critical to make sure that we as aura really stay close with our members on what they truly want and building features that are truly not just wanted and needed but also necessary. So that's another unique part of this launch. So interesting. Do you, you know, I hear, there's a couple things in there I want to get into deeper. But you know, you hear that phrase that you used a lot where it's women's health has been underserved. When you think about sort of, you know, a woman taking to the internet and searching things through an LLM, through a chatbot, through just a Google search, through any of these things, do you think that what does exactly underserved mean? Are women getting bad information? Are we getting a mix of information? Obviously, aura, like, tailors information based on other things going on in your body, which is awesome. But is there like a larger problem of what women are sort of finding and some of these chatbot results or Google searches? Well, I think we have to kind of take a step back and look at, you know, research over time. And we know historically that women have been underrepresented when it comes to a lot of research, even when you look at large pharmaceutical trials. We actually understand today that if the trials were really tested on women to the same amount and with the same level of depth as other participants, then potentially dosing might change. And efficacy might change of these drugs. And so taking into account -- Which is crazy. Completely crazy, right? Super crazy. So taking into account the constitution of the female body, how hormones may actually affect your physiology and your ability to have certain responses to medication, to certain diagnoses, to certain treatment plans, I think we see that kind of historically of being an area where women have been underserved. So you take that snapshot and you say, okay, so looking at the breadth of research that's out there out there in the internet, women have been searching for answers for just as long as the research has been done. And the answers that they're looking for are really disparate. They're scattered. They're on, you know, a niche Reddit forum or, you know, they're kind of word of mouth. They're kind of thinking about, oh, my grandmother and my aunt said this happened. And so because we're related, maybe this will happen to me. So a lot of it is hypotheses driven kind of data gathering, one off. And I really feel like we want to change that narrative when it comes to women's help. When you look at large, kind of search forums like Google, you and your training models based on everything, you're training on a lot of really amazing and rich and in-depth information and really reliable information. But you're also training on the flip side. You're also training on a lot of hearsay, making connections that might be causality. And so we really feel like if you can get down to the crux of picking and choosing the right training data, the right sources, the right guidelines for women's health, then you can start to push away some of that noise. Of course, we have a long way to go when it comes to the actual research. But you have to start somewhere and for us at aura, putting our female members first has been foundational for so many of the features and products we launched over the years. And so this is another reason why we wanted to take this approach. So basically with this LM, you were able to just select all of the information and put them in like a pool. And then the aura advisor basically only pulls from that wealth of information. Is that, do I understand that correctly? That's correct. An additional aspect of the process of building this LLM is, yes, the knowledge base is extremely important. But we also have trained the AI responses and fine-tuned with real clinicians in the loop in the process. So it's not enough for the AI to just be pulling from a certain knowledge base. But we really need clinicians to evaluate every output and say, okay, what I actually say at this way in the room, what is the structure and format, what's the tone, where does the empathy come in, what words do we use? Being really intentional about every aspect of the evaluation of the AI, I think has also been a critical differentiator. And it's really important for us at aura to make sure that we are keeping that clinician at the center of that process. So my role was really testing our AI, looking through thousands and thousands of AI output, giving feedback on what responses seemed aligned with what I would say in my clinic, or our director of women's health, Dr. Chris Curry. She's a practicing OBGYN, something that she actually would say to patients in her practice. We wanted to make sure that we were really building the AI and fine-tuning it based on this real world experience of clinicians today. And I think that's another layer that really helped us build a product that we can really stand by. Talk to me about the investment in this, because from how I understand it, this is the first wearable to have a custom proprietary LLM built for women. And I'm curious, how long did it take? How many people did you have to hire? What does this breath of research look like? That sort of filling it? Yeah, what did this all look like? So this process kind of started probably at the same time as we launched our advisor back in, I think it was March 2025. So about a year ago, we launched our advisor product. And we knew with that launch that the first area that we would want to build a more custom model would be in women's health, because we see such a need with our female members. So this process and this strategy was very much thought about at the same time that we were thinking about our flagship advisor product. In terms of investment, I think it really takes a village to build something like this. This is an extremely cross-functional effort. And I am just one piece of the many, many cross-functional teams that have been involved from AI, science, and R&D folks, to kind of the clinical team, to the product team, to the engineering team, the technical folks, all the way through product marketing. So every single vertical of aura has been touched by the build of this product. And I think it is definitely an example to us of that one team mindset of when we all come together with the idea that we need to do better for our female members and build something that is going to help them address their unique experience, their unique point of view, just all of the folks at WARA who came together to make that possible. So I long-winded the answer of saying, "Huge investment and will continue to be an investment in a focus for us as a company." And I feel really grateful that my role allows me to be so cross-platform and so strategic and work with such brilliant people. I learn something every single day from folks on our AI, science team, from folks on our kind of software engineering side, from product. It's been a highly kind of diverse and cross-functional experience. Wow. What was the testing process like? Did it take a lot of iterations to kind of get to the place where you're really confident in rolling it out? Sure. And the testing process is something that we will say will never be fully complete. We feel really good about the product that we've put up today, but based on feedback from members and labs, based on the questions that we hear from members. what they are asking advisor, we will take all of that into consideration and continuously test and optimize our product and the AI experience. So that is something that I believe will never be finished and we will continue to hope to improve just to meet our members where they're at. In terms of how long we tested until now, this project as I shared has been in the works for about a year now. And so it's taken many, many months and many iterations of testing thousands of Q&A outputs when you're engaging with an LLM like advisor, there are multi-turn conversations. So there's a level of complexity of trying to understand, you know, if a member asks X type of question and they get X type of output, what's the next question that's going to come along? And how can we make sure that we're, you know, building the right scenario around what members are asking about? So it's highly iterative, it's super complicated. Thankfully our AI science team, they are brilliant folks who are really building our platform infrastructure to do a lot of this. And so I feel really lucky to get to work with them and to build that alongside them. Here's something every brand and agency managing affiliate or creator programs is dealing with right now. You've got one tool for influencers, another for Amazon affiliates, maybe a third for publisher affiliates, separate dashboards, separate partner networks, separate reporting. It's expensive, it's time consuming and it's holding your team back. LaVonta is the affiliate and creator platform built for modern e-commerce. You can now run one unified program across Shopify, Amazon and Walmart all from a single platform. Discover high performing creators through an AI powered marketplace, automate product seating, set commissions or negotiate flat fee partnerships, and measure performance across every channel. So you know exactly what's incremental and what's building real growth. The brands that consolidate their creator and affiliate programs into one platform are saving time, cutting costs, and making smarter investment decisions. The ones still stitching together separate tools are falling behind. See what a unified creator and affiliate program looks like. Visit laVonta.io/glossy to get started. And right now when you book a demo, you'll get a $100 gift card of your choice. I am an or a user like I mentioned and I opted in. It's meant to be a pretty seamless opt in where it just kind of starts giving you new answers and advisor. And how are sort of the consumers meant to think about being part of this test and sort of where did the test results go. So the new model is available to test an or a labs and it's within the existing advisor experience. It's available on Android on iOS to any or a member who's enrolled in women's health features and all female members who have their app set to English. The way we think about this is we have a way to a B test in or a labs. So members select members make the opportunity to give us that feedback when two different types of responses pop up. They might be able to share which one they prefer and send us that feedback. So it's not all members who will get that option because for us we need to have a lot of different with with with user research. You need to have a control group. You need to have the testing group. You know, so there's a little bit of nuance with who gets what option in the labs experience. But for select members who do get the experience, they will be able to tell us a versus B which responds they would like it's blinded. And then we get that feedback on our side and we're able to see we're able to see what that looks like. We also will have opportunities to look at survey data. So from the or a labs experience with advisor when the custom women's health model is responding, we hope that members will click the survey button and will let us know kind of both quantitatively through the survey questions and qualitatively what's landing well, what needs improvement. And there are topics that they're really excited to use advisor for now when it comes to women's health. Where they may be seeing gaps in just other offerings and where or I can meet them with with their needs. We're really hoping for the ability to respond effectively with data for the entire life course, the entire female life course. Trying to conceive all the way through menopause. We hope that advisor can be that companion for our female members. And so we need feedback on what areas are we falling flat. Where should we strengthen the model. Where should we strengthen our knowledge base. So all of that is kind of part of that feedback process. And to tell you from first-hand experience, we have active channels that are talking about this every single day throughout the weekend. We are thinking through, you know, who's asking what type of question. Where are our gaps. Where should we be kind of building more kind of gray area tests of the platform. All of that. So this is kind of fresh, you know, coming off of a meeting, just talking about this before speaking to you. And what kind of data, what kind of studies are really fueling what the chatbot is talking about. Where did you guys get this data? So the knowledge sources were hand picked and vetted by our board certified practicing OBS. They are kind of all of the guidelines and research studies that are publicly available that our OBS feel like are the gold standard for how to build clinical responses. So really we relied on our in-house experts to tell us which sources to pull in. And it was, you know, in the magnitude of hundreds. And we feel really excited for the fact that we got someone to kind of really think through what is truly used in clinical practice today. And what kind of standards were the doctors looking at when they pulled in these hundreds of pieces and articles and sort of pieces of information? Yeah, in medicine, there are kind of publicly published standards by, you know, medical societies. And so the American College of Obstetrics and Gynecology, ACOG, they are the premier source for a lot of these. So we definitely made sure to follow ACOG standards and recommendations in building this model. Okay, and then talk to me about sort of this idea of like the chatbot having context context about your life context about your health as a user. So you're taking all of this vetted information. But then what does the nuance look like in terms of these new conversations and stored conversations, which I know can be very long in some cases between the chatbot and the user. So I think the way we like to think about it is advisor is your kind of always on companion or coach when you have any sort of wellness question. And you're looking at features like your sleep or your activity or your stress and you're saying, huh. What's going on at this phase in my cycle? Could this be related to how I'm sleeping? Could this be affecting my sleep or my sleep affecting my cycle? Like help me figure out where this data is and what you're seeing. And so what we're out will do an advisor does in this situation is, okay, the question is, you know, can you give me insights based on my cycle or, you know, based on my typical sleep patterns, how can I sleep better during the third trimester pregnancy? So the order data is pulling in your other parts of your app, right? So your sleep, your movement, your recovery. It's pulling in that data and it's saying, okay, given this context, here's what you can do. Here's some lifestyle modifications to kind of meet your question. And a critical part of this process is making sure that our members know that we as or do not diagnose medical conditions, we do not want to replace your healthcare provider. We don't provide emergency or individualized medical treatment advice, but we can tell you about your biometrics signals and maybe things that you haven't been noticing have been going on for the last one or two weeks. We can connect those trends to your question and put it in that context and then hopefully give you kind of an end of one snapshot on what's happening in my body. Now I understand kind of what's been going on. Here are some tips and suggestions that are data backed clinically valid to help you troubleshoot what to do next. And advisors always there to kind of be there when you try something and you come back and you say, okay, this worked in this way or this didn't work. What is my data show now? Can you help me track what's been going on since we last chat? How does aura prioritize old biometric data, old questions and sort of newer things? Are they sort of all like sort of like layered in there to create like the best result in these questions? With the advisor experience members have the control so they can see what conversations have made it to memories and they can kind of go back in and delete memories to kind of see what the memories are. And if those memories are kind of part of your profile, and if you have chosen to keep them, they will get pulled in to the context of the responses from advisor. But if you as a member feel like you want to remove certain memories or you know, delete it, then it's also up to the member to have that control. - How do you sort of envision the, can I say personality of the chatbot? Is that fair? How do you envision this? And how did you create it? - Sure. - Well, I think the aura advisor, tone of voice has been a very intentional strategy from the beginning when we launched advisor. And our tone of voice has always aimed to be extremely empathetic, extremely caring, thoughtful, intentional, not pushy, very much waiting for the member to share what they want and where they're going and to offer suggestions and partnership. And so when you take that, when you take our aura tone of voice, which I feel is a unique differentiator of our AI product and you put it into a more niche domain like women's health, how do we make sure we pull that tone of voice through? But when it comes to a lot of women's health questions, that the clinical aspects of those responses also need to be thought through. And so women are coming to us with questions and they want answers. And so can we think about how do you structure that response? How do you open, you know, with, you know, in my clinic, if a patient is coming to me and talking to me about something that's really personal, can feel awkward, can feel really vulnerable. How do you validate their kind of concerns and their experience with the opening line or the few sentences that you say at first? Then how do you root your kind of answer in research and data and kind of come back to the facts? And then how do you bridge that with kind of shared decision making? The shared decision making approach, I think is critical to what we are building at aura with our AI is making sure that whatever we are suggesting for our members really is something that they feel they can do or they're interested in. And we think about that in medicine all the time. You know, there might be, you know, two or three different options when it comes to a care plan. But the best option is going to be the one that your patient is actually going to follow through on and has the ability to invest in because that's really what it is. It's an investment in their time and their energy to make a change. And so we try to build that into our AI as well is making sure that our members feel like they are in the driver's seat, that we aren't telling them what to do. But rather they get to co-create the plan with the advisor. So that's a little bit about our approach. A lot of it pulls from what we learn in, you know, your medical training and wanting to make sure that that's pulled through into the AI as well. After this test is sort of rolled out fully and is become fully integrated, I'm assuming, into the advisor, will the advisor and theory talk to men and women differently? I think the way that I would think about it is our advisor is kind of this always-on companion in coach. And we want to make sure that we are meeting a differentiated need when the question warrants it. So when it's a women's health question that might require a little bit more, you know, vulnerability or, you know, it might require pulling in data from specific women's health aspects of the app, that is going to feel like a slightly different experience than, you know, a non-female member. However, our goal is to pull in that relevant data to have the same kind of empathetic tone, to have that clinical validity across the app for all of our features. And so we've built something kind of proprietary when it comes to the knowledge base, but the external output of how advisor engages with every single member that should feel, you know, unified throughout the platform. So we don't think about this as being building a separate experience, but rather one unified experience. But just for the women's health questions with our current model, when the women's health question comes in, you can believe that the knowledge base that's being pulled from is really clinically validated for women's health and current guidelines. One thing that I feel like might potentially be, I guess like a, you know, a struggle for, for aura or for anyone else who does this, is this idea of, do you think that consumers want a diagnosis and then they might potentially be disappointed when they don't get a diagnosis? And I'm asking that from sort of the person 15 years ago that would go on WebMD and like look something up and want an answer. And I think that consumers still are sort of acting like this. But of course, we're a cannot diagnose and neither can other chatbots. So where's the sort of expectation from consumers and all of this? I think we're entering a world where you see, I think there was a recent report and, um, like, see, I do need to kind of verify this. But I think there's a recent report that eight, eight out of 10 consumers are kind of going online for their medical questions every day. Something, I believe it. Something like that. I believe it. I think we see that there is so much information out there and it's accessible at your fingertips, the press of a button. And you see it all. And I think there's an expectation that because there's so much information, it should be easier to come to an answer or to a diagnosis. That also should be faster. And I think that the caution that I have as a clinician is, I'm actually so excited that our patients have the ability to come in so equipped with knowledge to their clinic visit. But there still needs to be someone who ultimately can see the full picture has the medical training, can build the trade-offs, can think through the shared decision making with the patient before they ultimately decide on a diagnosis and treatment plan and all of that. And those are really big considerations. So I would rather say that in this day and age, even though the need, really, the pull that we're hearing from consumers is we want answers now. I would say it's really beautiful that our members and patients and folks who go online can get so much information and can come in with so much understanding about their bodies. Where they're really taking control about what's going on in their individual physiology. But it's still as important to have that human in the loop, that physician in the loop, to ultimately make those diagnosis decisions, those treatment decisions, because that's what they've been trained for. That's what they've gone to school for. But when it comes to education and understanding and context building, this process of going online and coming equipped with so much information can actually help your physician understand what's been going on, for those months or weeks that have gone in between clinic visits. It can get them up to speed so that they can make decisions with you so much more precisely without you having to repeat yourself, without them having to waste time on asking questions that you already have the answers to or your data already shows. I think that's what's really exciting about this new paradigm of care is getting to enable that process, where you unlock that clinician to really get to do the job that they're really excited to do and train to do. As a patient, you come in with such a deep understanding of your body that you're able to enter that shared decision making process so much quicker. I think that's the beauty of what we see going on today. Yeah. Do you think that there's a sort of inherent risk of people sort of over-relying on a chatbot? And I asked that, there was a study in Nature Medicine Journal, and it found that chatGPT health under-triaged over half of questions around emergency care, and it basically should have sent people to the emergency room, and it didn't. And I'm curious, when does Orra know to send someone to the doctor? When does it know to kind of stop and say, like, this is not a question for us? These are critical edge cases that we, as the clinical team, have kind of been building those safety frameworks around from the beginning. So we actively test our advisor product on these escalations and want to make sure that we're really staying really tight on when things do need to get escalated to an urgent care or a 9-1-1 that we aren't just continuing a conversation when there is a critical need on the members behalf. So this is a constant kind of process that we are optimizing and building at Orra today, but also we have always taken a very conservative stance when it comes to things like that, because we want to make sure that our members' safety and trust comes above all else. There are also other concerns, hallucinating, drifting, privacy concerns. How do you sort of present these kind of concerns to consumers? How do you sort of position things for them? You know, a huge question. And I think what you're alluding to is kind of what happens if our models hallucinate. What if they give the wrong answer? What if they give a harmful answer? How do we make sure that we're detecting this? How do we make sure we're fixing this? So this model has been introduced as an experiment in ORLABS specifically for this so that we can rigorously evaluate quality, safety, member satisfaction before we do any or consider any sort of broader rollout. As I mentioned earlier in our conversation, the clinical team, we are engaged every single day since this has been rolled out to monitor feedback, engagement patterns. Any flagged responses? Are there areas where the model means refinement? We are kind of on call for things like that. And because this is our own model, because it's been curated on a clinician vetted set of knowledge and because it's operating within ORLA's kind of privacy first framework, we have a tight control over the outputs then with a general purpose kind of system. So when these issues are identified and there will be issues and that's okay, that's part of the process of building tools in AI. We have the ability now because this is ORLA-owned to quickly update prompts to improve the guard rails to test that the underlying knowledge sources and make sure that we are keeping accuracy and safety as kind of being the table stakes for whatever experience we build for our members. Tanya, I am so appreciative for you giving us all of this great insight into here. I'm curious what you think the next year is going to look like the next five years, the next 10 years. What do all these themes that we're talking about turn into eventually? That is a great question and I am in a unique position where I see not just what we are thinking about at ORLA and our AI strategy today, but where we want it to go. And you can see that ORLA has been very intentionally investing in clinical care and making sure that we are building the right infrastructure to address a world in which ORLA is thought of not just as a consumer kind of ring wearable company, but as a true healthcare company. So we see our launch of this custom women's health model as part of that broader effort to thoughtfully explore how to advance that member experience. There could be a range of approaches. We are thinking about them all and we're just focused on building. We're focused on partnering with the right companies, the right health systems, partnering with clinicians and patients to build capabilities within the ORLA experience that's accurate, personalized, actionable when it comes to health insights. And so this space has already evolved so much. It's going to continue to evolve and our strategy will evolve as well. And I think it all comes down to just being really innovative but also thoughtful and being kind of safety first when it comes to our members and always kind of putting their needs and their feedback as being paramount and as the North start that we're building towards. So whatever we hear from members is kind of what dictates what we build. And I think it feels really gratifying to be at a company like that. And in medicine we think about being patient first and I think for me being patient first at ORLA really translates to being member first. As a user I'm always thinking about sort of what will all of this just feel so archaic in a year like things are moving so quickly. It's so hard as a consumer to know what we're sort of expecting. But do you think that the whole world of how consumers will use apps, will use trackers, will use wearables. And everything will sort of be changing really dramatically. I think we see new launches releases, capabilities coming out on the market every week. And that's really exciting from someone who is kind of building in this space because you feel like you are so motivated by your peers to really build something that's the best for your members. And so there's a lot of motivation that really is pushing the space ahead. What I'm really excited about is how longitudinal kind of signals like your biometrics really layer in to the personalization that you're able to achieve with AI. And I think that's a place where wearables are truly differentiated because that's a knowledge of an individual member that is so unique and rooted over years of history. And so the predictive abilities of AI or the early detection abilities of AI for an individual, I think that is what is, at Stanford we talked a lot about precision medicine. And I think that it felt like something that was very far into the future. But I think with the advent of so much of this technology and the AI space moving so quickly, I'm actually really excited about our ability to truly achieve precision medicine with something like this. And so we're on all of these different data types. And I think that's where we're really going and we're going at kind of break next speed there. And it was something that I did not think was achievable in the near term five years ago. But here we are. >> It feels like someone who covers this industry, it feels like everybody is working on things right now as quickly as they can. And then one person releases something and then it's like, oh, we got to release. And it just feels like this flood of stuff that's happening right now. It feels very exciting and it also feels like, wow, we're kind of on the sort of turning point of how AI is going to be used for all this stuff. >> Sure, yeah. That's how we feel at WARA as well. We are growing quickly to meet these needs and to make sure that we're building quickly and are being agile to meet our members where they're at and to meet them with what they're asking for. So we similarly feel that internally. >> Hanvi, thank you so much for chatting with us today. So much amazing information and yeah, thanks so much for coming on. >> Thank you so much for having me, Lexi. And thank you for listening to this episode of the Glossy Beauty Podcast. If you enjoyed it, please leave us a rating and a review on Apple podcasts, Spotify, or wherever you're listening. And for even more coverage on the beauty industry and more, please visit glossy.co/beauty.