Name: 8.21_On-Demand [edited]
Uploaded: 2025-08-21T22:17:41.069Z
Duration: 52 min 57 s
Description: 8.21_On-Demand [edited]

Transcript for "8.21_On-Demand [edited]": Webb and I are delighted, to be here today. The two of us have been on this incredible ten year journey of Komodo, and, actually, we're hitting our eleventh here very soon. We're super excited to to welcome all of you here. Webb, please go ahead. I'm gonna jump in. Alright. Alright. Welcome. Most of you know me and Arif or one of us. For those of you that don't, I'm Webb. I'm the president and cofounder of Komodo Health, joined today by Arif, our cofounder and CEO. The two of us welcome you all to today's webinar. We just wanted to say thank you all for your interest and for joining us for this exciting session. Arif and I are so excited that this is the first new product announcement that the two of us have launched together into the market since we started Comodo over ten years ago. So why are we all here today? We're here because at Comodo, we believe the most profound insights to drive your business, to improve patient outcomes are still trapped in the complexity of health care data. And we believe that we're at a critical inflection point in terms of the unlock. Our mission at Comodo has been to build the operating system that liberates those insights and empowers all of you, the experts, to move our industry forward so that we can collectively reduce disease burden. Our full stack vertical thesis has been in place since Komodo's inception over ten years ago. We've designed, architected, and built out a unified platform and experience where data, technology, and health care analytics live in concert. It's one thing to piecemeal together solutions from disparate vendors with different capabilities, but many of them are gonna lack either data, analytics, or AI experience and expertise, which is why it's an entirely different experience to partner with a company that has thought through how every layer, every team, every discipline, every dependency works together to power insights that can impact your business across the entirety of the product life cycle. This isn't just an a design or an architectural choice. It's our foundational belief that true insights cannot exist in a fragmented world or ecosystem. Every layer of the full stack thesis from our healthcare map, to our platform, to our applications, to our AI is designed to work seamlessly as one. But before we go there, let's talk about AI. It's hard to miss today's discourse around AI. AI is everywhere. It's reshaping every industry. The real story to us is that AI is amplifying human potential. There's examples everywhere. It's expanding creativity for artists. It's making athletes smarter and stronger and healthier. It's turning language learning into a conversation. The pattern to us is becoming very clear today. Today, AI isn't replacing human expertise. It's removing the friction and giving users superpowers. I'll I'll give you a couple specific examples. RF knows. I love music. I love basketball. I love, sneakers. And so we'll use those as as examples. Think about Grimes and AI. Grimes' fans are generating new songs using her voice, using AI. They get to cocreate with Grimes. They collectively earn royalties. To us, that's expanding, not replacing artistry. NBA teams, they're using AI powered wearables and AI powered video analysis to fine tune movement and prevent overuse injuries. That's super cool. You think about the Nike Air collection. Athletes are dreaming up bold sneaker ideas, and AI is helping bring them to life. Designers get to design and refine. Athletes get to test, and fans even get to create their own custom shoes all using AI. That's expanding creativity and personalization, not replacing designers. Those are just a couple examples that, you know, I personally found compelling. But when we look at it, there's one industry where this promise still feels out of reach. And it's our industry. As we all know, health care isn't like every other industry and health care analytics isn't like every other capability. The stakes, the data, the trust, and the transparency bar are all so much higher. Our collective impact is measured in patient outcomes, and the margin for error is zero. The truth is ours is an industry with data that's inherently complex and uniquely fragmented. It encompasses everything from ICD 10 codes and LOINC standards and everything in between. It's about intricate relationships between diagnoses and outcomes. It's about a chain of trust that runs from the clinic to the boardroom, and that's really where general purpose AI falls apart. And by the way, it's not just us saying this. We're hearing this from all of you every day. When we talk to health care and health tech leaders and experts, we hear the same questions, the same frustrations. To us, that's not a trend. It's a signal. And this is what that signal sounds like. It sounds like people to coming to us saying, AI is powerful, but where's the business impact been in our business? My OpenAI investment isn't giving me anything I can act on. We're still waiting weeks for answers. Why is that? And you can read the rest of the the, signals that we hear every single day. What that tells us is AI is pervasive, but results that you can trust and tangible business value and outcomes that you can realize are still out of reach. This is the problem, and this is the moment that Komodo has been building for. So what's holding everyone back? Because we know your investors, your c suite, they're asking you the same questions every day. What's our company position on AI? What are we doing with AI right now? What can we be doing with AI, and why haven't we done more already? The questions are coming at you every single day. The truth is the nuance within health care is unique. Horizontal AI lacks the experience and the context to present a solution that understands healthcare. It's why generic AI struggles to comprehend and process healthcare's complexity. Now on the flip side, legacy platforms, your traditional software partners, they require extensive technical resource. They require differentiated investment into AI expertise. And for you, partnering with them, that's creating bottlenecks and offering a limited experience. This is what we all need to collectively overcome. And here's the reality. Generic AI models and generic horizontal large language models are just not built for this moment. They're trained on broad, non clinical data. They hallucinate when the when the context gets complex, and they can't meet the evidentiary standards that health care demands that we demand. And so we ask ourselves, what if AI did speak the language of health care? What if AI understood that context, understood those nuances, and the fact that unique advantage can be realized when AI is engineered from the ground up with healthcare in mind, not as a bolt on, not as an iteration off of old pre existing approaches? The answer to the waiting, the answer to the context tax is a new kind of intelligence. It's an intelligence that has domain expertise to complement, enable, and accelerate your abilities. It has to think in health care and be powered by AI ready data. And so with that, I'm gonna pass it over now to the star of the show, Marmot, featuring Komodo Health CEO, our very own doctor Arif Nafu. Thanks, Webb. And thank you everyone for being on today. You know, one of the most amazing things about, this business and and where we started Komodo was this idea that, we could take, an incredible amount of data and we could drive to these incredible outcomes. And, by doing that, we would be able to make AI, or make our data and insights available to everyone. And so today, we're excited to introduce Marmot to the world. We we believe that Marmot really is the embodiment of eleven years of Komodo. And what we mean by that is that, for so many of you who know us, you know us because we've got this incredible real world evidence dataset, or maybe you know us because you use our software like Prism or Sentinel or Aperture or Iris every day. And for those of you who know us either through our data or through our software, imagine if you were able to bring AI to that entire conversation. You could have a conversation with your data, if you were able to actually bring your expertise to bear because of AI. And that's really what Marmot is designed for. It's designed to enable and empower you as the health care experts, as the analysts, to have a true copilot that's gonna help you get to your answer better, quicker, and more reliably. And we do this not by shoving a ton of data into the AI and having it process it. We do this by actually teaching the AI to be your code partner, to be your thought partner. And this is what's leading us to this incredible product experience and this incredible outcome, and we can't wait to show it to you today. Yesterday, we had the formal announcement of Marmot into the market. And it's a very, beautiful day for us because it's an opportunity to talk about impact. We're really delighted at this launch to name one of our first partners, which is Alnylam. Nylums is this incredible company that has brought meaningful innovation across a number of disease categories, and we at Komodo couldn't be more thrilled to partner with them. We're building actually on a five plus year journey with them. And now Lytle has selected Marmot to bring AI to a number of workflows. It's really been a delight to see how their business thinks in a very forward thinking way about the way that AI can transform their business and drive better outcomes for patients. And this is really what makes Martinip special. It's not just an outstanding agent built on Komodo's data to answer questions on the Komodo healthcare map. It's also an AI platform that will allow you to get control over your own data and your own workflows. In fact, what we're showing you today is built entirely on Komodo's technology. As Webb said, it's using the best of foundational models, which continue to get better and better and better, but bringing this incredible context on both health care and what it means to be a health care analyst. For those of you who don't know either Webb or I, the two of us started as consultants. We did the job, working with commercial operations teams, HEOR, medical affairs, clinical development. And Komodo was born out of the passion of two analysts who loved the work, but believed there was always a better way, a way that we could use technology against better and better data to get better results. And that's what we're really showing you today and what we believe the ultimate value proposition is of AI. It is designed to drive better outcomes faster, better, and more effectively in the hands of our greatest users. And our users, all of you, healthcare analysts, folks who are in charge of brands, folks who are in charge of insights, folks who are running studies, the idea is how do we give you the tools you need in order to do your jobs better? And that really is what motivated us to start Comodo. It's what's really motivating us to release Marmot to market. Related to this, we always see, an experience, an experience that starts really with a single question. And for us, that question is, what can I do to help? And this is the same question that, you know, our clients come to us asking or we come to them saying, hey. Well, we we'd love to help you. What can we do? And so Marvin starts with that experience. And what I'm gonna show you today is how we actually bring this generative experience into something that's really cohesive and designed for all of you. And so with that, I'm gonna go ahead and drop the screen share and actually turn to to Marmot. K. So while this is coming up great. Okay. You see a lot of words on your screen, and I'm gonna spend a little bit of time describing what you see. So when you log in to Marmot, this is the product experience. On the left hand side, you have this full idea of seeing the analysis that you are running. And I have a number of analysis that I'm seeing right here, different kinds of example analysis I've pulled up for us to talk about. In the middle, your eyes will go to that naturally. And this is the place that we actually have a conversation with our data, and have a conversation with Marmot to actually pull out the things that we wanna see and perform an analysis for us. And then on the right hand side, this is probably the most important, tab for you, which is where what we call the code panel. This is where you actually build trust with the AI. And you will see something incredible today, which is that instead of the AI just processing your data and returning a result, it's gonna be incredibly transparent about every aspect of the process. When we talk to our customers about the number one reason they are having struggling with foundational models tied to their data or large projects that are that are focused on AI, the the biggest problem they cite is trust. They can't really trust the result because it's not transparent how that result was obtained. So we're gonna start with an example of a question that our customers ask us. We get these questions every day. This one happens to be on ALK inhibitors and, duration of therapy and a kind of a comparative analysis between this and other ALK inhibitors. And and and this is where we really start here. We start with this kind of question at the top. I'm just asking it. And the first thing that, Marmot offers is this notion of a research planner. And this research planner is asking you to clarify your preferences. So you've got this clarification, but it starts by just generating an analysis plan. It's kinda what you do as a health care analyst. You start with an analysis plan. Here, I'm gonna conduct a kind of comprehensive analysis of treatment patterns, one drug compared to other ALK inhibitors. I'm gonna start by defining the cohort. I'm gonna define my outcome measures. I'm gonna define my dose modifications, discontinuation patterns, FOXI measures for discontinuation reasons because we're using claims, for example, and potentially lab results. And I'm gonna talk sort of describe the statistical analysis I'm gonna run, and I'm gonna describe the sensitivities I'll run. Now before you start an analysis like this, you will often say, well, there's so much art. I always say the the work of an analyst is an exercise in art. You make this artistic decisions every day. How do you define a therapy? How do you define, duration of therapy? You know, the number of days that I choose to use in the index against my index period, these are all autistic decisions. And there are many different standards you can choose. And so our AI really should be able to ask us the most important things. How should we define initiation of Lorebretta? What allowable gap between prescriptions? How should we think about discontinuation? How do we identify dose reductions in claims data? Do we just, you know, observe the claim event, or are there other proxies for this? Which comparison approach should we use? And so what, what Marvin allows you to do is it not only asks you these questions, it gives you what it would believe are the most common answers to these questions. And this is really wonderful for the analyst because you're just gonna see this and intuitively know. No. I wanna do this method or that method. And maybe I wasn't explicit about it, but my AI's job is to kind of make some of these implicit decisions explicit so that you can tell it and instruct it to follow a pattern that you like. Now while you're doing all of this, this pool panel is starting to populate. First thing it's gonna do is tell you about potential pitfalls. Like, what's the challenge with doing this analysis? Well, you're gonna have problems with interpretability around discontinuation. Again, if your data is not closed or continuous laying rolled for a period of time, maybe you're gonna see different reasons for discontinuation. So I really like this because we built a full peer review process into this where it simulates the idea of you getting feedback from someone else. But the nice thing is that you're having this conversation with your AI to do this. So now it's gonna give you pit simulation this thing, you know, give you kind of pitfalls for doing the analysis. Here's another point, data latency issues. We understand that the observational windows, from a latency standpoint has a huge impact on the outcomes, which you can actually model and ask Larmen to model for you, but it's right there. So it's gonna get started on this analysis I clarified, and it's gonna start to run it here. And so this is an example of it running through the process. Now what's happening on the right side of this panel is that it's actually doing the work in a very transparent way. It's defining NSCLC. It hasn't have a definition. Well, I'm using claims. So we know there's gonna be some definitional issues from a proxy standpoint. It's defining the different drugs, and it's telling you what it's doing and why. So it's saying, well, for this drug, I've selected this code set, which includes the NDC specific for it for the oral formulations, and it's the most appropriate codes to identify the patients. It also tells me what I'm excluding. So when I make a decision like this, I can see what the AI is doing. This is so important for a lot of our customers because they're like, hey. We wanna make sure that we have control over defining the code sets, defining the methodologies, and guess what? The AI is doing that. So once it defines all of the codes, this is the part that's really awesome. It generates the code for doing the work itself. But before you get to this median treatment persistence by Alconybitor, you are actually seeing the code that it ran on our data. You can take this. You can run this in MATLAB Enterprise. You can bring the actual code to the high code environment, and you could do it yourself. So that reproducibility for us was really paramount to us solving this problem. And the problem is not that an AI is not great at generating code. The problem is that AI is making mistakes all the time against both intent as well as just the intricacies of the code that has to be generated, whether it's SQL, Python, R, for the problem at hand. You spend so much time QA ing the code to make sure that the intent of your work is exhibited in the code, and that's what it's doing here. It's actually telling you exactly what's going on. It's showing you how it's performing the analysis. It's giving you transportable code that you can run on our data, and that's really important to building the trust. Now what's great is that after it generates this, it's going to start doing this. We did this for one drug. It's now doing it for all of the inhibitors. And it then creates the cohort for the next inhibitor, generates it. And so methodologically, it's reproducing the method across each of the drugs, which is awesome. And then it's actually gonna prepare the summary looking at kind of the characteristics. You can sort of see here around the cohort, age, count, so on and so forth. So it's generating kind of preview stats here of what it's doing. It's also self regulating. It's figuring out what it's doing wrong. It's fixing it. So, it's going through these steps one at a time, and it's bringing that to bear. And then, of course, it's going to once it does all of this, it starts to generate the results. And here's what you see in the middle, the results being generated. You can see these dose modifications across. You can see the incidence of, all of of key events or adverse events by all contributor and percentage of infected patients, and it has a definition of how to find an adverse event. And this is all transparent to you. So you're like, look. I I I see these AEs, but, you know, I'm gonna define cardiac arrhythmias with these other codes. You can do that. And the beautiful thing about having the conversation is that you're seeing you're playing this out in real time with the way that you think about the problem, and you see it being pulled through. So this is really what Marmot's doing. It's going through that whole process. And then, of course, at the end, it's summarizing and giving you a set of limitations, for the, study. And then it's also kind of giving, a response to those limitations. So this is something that we call our independent review agent, which is super exciting. It kinda comes in and does an independent review of your methodology and outcomes, but we make all of that really available to you to run. So this is the structure and this is the component of of Marmot. There are so many ways to get, to use the system. You can ask questions about data quality and fidelity. You can ask questions from a commercial audience point of view, medical, clinical development, any part of the pharma life cycle where you're doing work on patient level data, projected data, your own datasets, Marmot's really there to be that, place of performing analysis. I'm actually gonna show you what this looks like for a very simple use case. And and the simple use case we picked, we we take a question we get all the time from our commercial audiences, and they say, hey. Help us, you know, defile these prescribers. Now this is a a problem that's, you know, been in the market for, like, forty years. We all need to know who commercially we need to go talk to, and we all have like, there have been twenty five years of standard methodologies of deciling, so there's nothing that's specific about it. On the other hand, we all have different ways potentially of doing some of the intricacies of the analysis. And so I'm gonna actually show you how how Marmot handles that, process and the kind of control you can have over the way the AI will follow instructions that you create. So I'm gonna just sip a simple question in here. I'd like to see the deciles for ATPs prescribing Kisali in 2024. So I'm expecting a deciling model based off of this. Now you can imagine deciling is done either by kind of taking the total prescription volume and dividing it up into tenths and then looking at the number of prescribers or saying I want equal prescriber bases. And so you're gonna see the AI pick a method that it believes is most correct. And, actually, we specify the method, and I'm gonna show you how to do that in the project prompt. First thing that comes up here is that you'll see this task list here at the bottom. It's gonna say, first, I'm gonna search for the Kastali codes, then I'm gonna create a data filter for Kastali prescriptions, then I'm gonna analyze provider prescribing patterns, Then I'll actually generate the deciles and visualize the results. And you now start to see the system doing the work. First thing, it's actually picking the code sets right here, then it's actually picking, the definitions of the Kisvali. And here, it's actually using both Kaskali and the Kaskali from our Copap codes, but it's excluding other CDK four sixes. Obviously, I didn't ask about Ibrance or Vicenia. So it's making the right decisions on picking the right codes that should be obvious. And it just basically said, now I'm done with the data filter. And now I'm actually gonna go and generate the query. Okay. So here's the SQL query to actually analyze the prescribing pattern. So this is the code that it's writing, and it describes how it's doing. It's using something called prescriber MPI or primary ACP MPI. This is actually Komodo MPI information in in Plaid, and you're actually seeing it perform the computation here. And it's writing the case of when the sum of TRX is total times first time percentage. So it's now starting to do the deciling in the code. It's ranking the prescribers, and then it's pulling together a finalized output here. So this is the step that it does. And here, it's actually starting to do the deciling. Now it's actually moving on to doing the visualization of the first analytic output, which is essentially the prescriber dispute distribution by decile. So you're gonna see the AI kind of run through this process of doing it, of generating the visualization. And what's really nice about these visualizations is that, you get the underlying data table behind it, and then it's just rendered into a nice chart. And, we will actually be introducing a very cool canvas mode soon where you can actually tweak the visuals, tweak the design, change the bars around, all with AI, all very quickly. But here, it's actually generated a nice distribution of providers by decile, and it sort of identifies, you know, how many prescribers are in each decile. So about, you know, the top 80 prescribers, you know, are are are re results in about 10% of the total prescriptions in this market. And you can sort of see how that curve works. This is a pretty standard curve for this type of oncology, oral oncology product. And so now it's telling me here that it's it's gonna create another analytic output that virtualizes the market coverage efficiency. So, like, you know, what percent of doctors would we have to see in order to get, you know, 50 or 80 or 90% of total prescription. So this is the kind of common work you do with, you know, in a death filing project. It's actually running through the steps. It's generating individuals. We're gonna wait a minute for it to actually come up with the second, chart here, and then I'm gonna actually ask it to provide me a summary. So here, it's like to capture 10%, only about 80 prescribers. If you capture about, you know, 80%, you need, 2,400 prescribers. So this is exactly the kind of work we wanna do. Very predictable outcome because it's a standard deciling curve that we wanna see, and this is almost a cumulative function of how many doctors you have to see to get whatever, prescribing. And so I really like this summary because it's sort of telling me the concentration, describing the disparity, and it's got a ton of follow-up questions. You know, what are the specialty and practice characteristics at the highest volumes? You know, you could you could sort of ask these. You know, is there geographic concentration? Are the patient profile difference between high volume and low lying prescribers? So all of this is available to you. And one of the things that I love about, an answer like this is that I can really tune the answer the way I am. So I can just ask a question like this openly, or I can give it an instruction set. And, for this, example in in particular, before this meeting, I actually set a set of instructions. I said, you know, I want you to when we do deskiling work, this is what I want. And this is amazing because you can do this. You can write your own project example of exactly what you want the system to do, or you can trust Komodo, and you can trust Marmot to make that decision itself. And so when you ask this question to Marmot, it will make all the artistic decisions. It will decide how to kind of decile providers. It'll decide how to visualize the data. It makes all those decisions itself. But the powerful thing is that you could say, great. I'll let Marmot choose, or you could say, I'll choose. And that's really what's amazing about the system. It's really designed for you to be that that choice. Now some of you are looking at this and saying, this is great, but this is kind of an easy problem. What about a hard problem? What about something that I really don't know the answer to? And so this is really where, like, the approach Marmot's taking shines. I think it's something hard. Like, using claims data, seeing the line of therapy of HR positive HER2 negative breast cancer is is a hard problem. And seeing the utilization rate of CDK four six is a hard problem. It's hard because finding the right window, finding the right way to define the population and heavily sample data where you don't or may not have very clear mutation data is a difficult problem. And what you really want is you want a thought partner. And what you'll see here and this this is a very detailed discussion, and we we have this incredibly intense discussion with this AI. And I actually go through it, a lot of different things on the ways to define the population and cohort, their difficult calls. And one of the things that it realizes is that, look, I'm not seeing I'm seeing a lot of ET alone. I'm not seeing as much CDK four six as I would expect in publisher. I mean, we actually go through a discussion, to to understand that. I was like, I'm surprised about the rates. You know? Let's make adjustments. This that's whether the drop is covered by our rules. Maybe our business rules stink. And by the way, if you have you know, like, we have clients that have a 100 page business rules in a PDF, a CSV, a doc. You can literally upload them into Marmot, and they'll follow your rules. They'll define your therapies the way you want it. That's actually how, we discover issues in our rules. We we upload that, then we tell Marmot. We're like, hey. Find mistakes in this. And it's like, oh, yes. Here are, like, five rules that are, like, completely ruining your outcomes. In a complex query, that's really important. It's like a wonderful way to actually use the system to generate, to test, and to improve your outcomes. We still see the e t plus, the CDK four six, therapies be be far below what we expect to market. And this is actually where you're gonna want mutation data. You're gonna want that to verify these patients. You're gonna wanna bring that in. But look at these Kaplan Meier tTNTs. Like, these are the exact kind of therapies I'd expect. Like, a CDK four six has the next treatment far better than ET alone, and that's what you should be seeing. So you can do a lot of validation on the outcomes. You can do a lot of validation on the methods for really, really hard problems. And I would describe this on the harder end of the kinds of problems that you try to solve with great proxies and claims data. And, again, all the code, all the approaches, everything is provided here to kinda get to the results. So that's one of the reasons that we love this product. It is the way that you can truly have a conversation on hard problems, on easy problems, transparent methodologies, specify them yourself, or work with the AI to actually ask questions and push. This is a great opportunity for you to do this work yourself and to empower every one of you with something that, like, you can actually use. So we're we're delighted to to show this to you, to bring this to life, to show you live examples. One of the things that Webb and I take a lot of pride in is showing you things that you can actually license right away. Our team has been using Marmot for months, for months. Anytime you guys have asked us a question, it is likely that someone pecked with Marmot. And every single one of our analysts vouched for the answers they give you back. So they own that answer. They come back to you with an answer. But this is their thought partner. And so as we were seeing them use this, both Webb and I were like, it is obvious to us that every analyst, every commercial analyst, every HEOR, like, scientist, data scientist that's doing really hard work every day deserves an analyst that they can work with. And so that's really what motivated us to get this done and to bring this to life and to bring this to market. So I'm gonna switch back to the slides. I wanted to share a couple more things with you, before we move to q and a because I think it'll help you really understand what's powerful about, this whole system and what makes this amazing. So okay. So number one, the only limitation is really you. I was thinking about this as someone who doesn't like coding anymore. I grew up, like, writing a lot of the, the algorithms for Compose first product aperture. I, like, learned Python to do it. Twenty years ago, I was I learned, like, MATLAB to to do real work. I don't write code now. And it's actually one of the things that has I've observed is that not only has the world started to abstract code away, but, like, the skill of a great analyst is the ability to ask questions. It's the ability to challenge the results they're seeing. It's the ability to to push when they see something they don't like. Now think about your workflow today. You have a somebody, CEO, ask a question, filters down, you get the question, then you're like, You go you call your favorite consultants. They come back maybe a couple weeks later. And by that point, the question is invalid. And we think about how do we close the loop to that insight faster? How do we give you the tools to just keep asking better and better questions? To us, this is really the power that this unlocks. It's that, it's that, like, ability to think and the ability to push becomes the language, becomes the way that you talk. And it was it was, I think it was Andre Capati, like, with the guy who developed the word vibe coding. Like, two years ago, they said, like, English is the next coding language. It absolutely is. And even for analysts, English becomes the next coding language. It is how do you push, you know, the methodology harder? How do you ensure that it's giving you rigorous answers? Well, here now we can generate the code. Like, we can revise the code because you can ask it the questions you want. So we really love that. What we think makes Marmot different as well, and we've seen, you know, dozens of solutions on market, we see that every company has, you know, company name and AI. Like, we won't name them. You all see them. They all throw out something and, you know, with the word AI in it, and they think they have something. Problem with that, for a lot of us, is that to actually use this in your day to day workflow, it's it's gotta have certain things. And this is why we invest in this. Number one, you know, the underlying data of Komodo, our health care map is really second to none in quality, in completion, in validity. Every month when we talk to the market about our data, we give you coverage ratios of our pharmacy coverage, our medical coverage to market. Our data has now been used for, you know, hundreds, publications, poster presentations, and so on. We have tons of market validation, and we're just getting started. Our, systems are now processing data every day, and Marmot's gonna have access to daily updated data in a way that you will be able to get instantaneous insights. Really exciting for us. And it's also changing the market from a quarterly to a monthly to a weekly. Well, what if it's just real time? Like, why can't it be? So we get to ask those hard questions. We get to do those hard questions, and that that brings the joy, that smile to us every day. The second thing is that we've seen, you know, hundreds of thousands, well over a million queries. People use Prism and Sentinel every time. We see every combination of offset filters and, these, sort of demographic features that I want and to build performance systems off of, like, very, very varied queries is really how we taught Marmot how to think like an analyst. And and this is one of the best benefits of working with us is that every day we're just adding new capabilities, new features that make this, the searching, the engagement much very easy much easier, better. And so that's really the best, part of it. The third thing is that it really understands your questions, and it provides context you didn't know to ask. I feel like I'm a fairly experienced health care analyst. I learn something every day. It it, it pushes me on something. It finds an assumption that I made. It challenges the way I think about something, and we all need that. Like and it's great to do it in the privacy of your own room because you don't have to sound like an idiot to everyone else. You can just kind of look at it yourself. You're like, oh, like, okay. Yeah. Yeah. I forgot to say that. It's okay. Like, it happens to all of us every day, and I think having an AI that you can work with where you get to to be the kind of person that gets to ask all your questions, it's like a very beautiful thing. And I think it's a very powerful thing, and the idea is to make you more powerful. And then finally, it's really already in production for us. We already use this, to help all of our teams. And we'd encourage you to think about what are questions that are hard to answer, what are or things that are difficult for you to get insights on, and and bring those bring those to Marmot. Because we we see this great opportunity to completely transform that workflow. And so, you know, related to that and and this is, you know, getting back to kind of what we were talking about. I had a CEO of a company ask me recently, like, where does all of this go three to five years from now? And I my feeling is that everyone's gonna have their own analytic, assistant that meets them where they're at. You're a CEO. You need to know next week's forecast. You're an analyst. You need to know five reasons that the field is gonna yell at you because IC computations are off. You're h an HEOR, and you need to have three ideas that are compliant that you can actually pursue a study on. There are so many things that each of us want, but we we will be counting on our AIs to surface these to us, to share what we need to know when, and to help us through the process. What we think about with Marmot are two levels. You have a professional level where you just use Komodo's base agent. That agent is your agent. You can upload files. You can ask questions. You can work with it, stream out the most health care map with all of our context. And we also had an enterprise version where you can actually build custom agent workflows. We have a strong evaluation bench. We have a strong agent builder. You can link whatever tools. We have we have a bunch of tools that we've built that you can start to build more tools, and then you can build processes for things that are either covered today or not. You wanna get a case report into the hands of a field reimbursement rep every Monday, you can do that. You want a systems, health care system report that you want in the hands of the field tomorrow, you can do that. It opens the door for you to be able to do things that previously would have been difficult that you can actually materialize with AI very quickly. So this is why the next era of health care analytics, AI built, AI driven, controlled by you, this is really where the power comes in and why we're so excited to bring this to you today and why we're so we are such believers in this that we, as a company, could have stuck with all of our solutions and never stepped into the AI fray and never just we could have made a Komodo AI, and you could have gone out and tried to, like, link to it. And it would've been a terrible experience, but we could actually bring you the next generation now, stand behind the product, work with you on it, and share something that I believe will actually transform your workflow. And we chose the latter path. It's the harder path, but we believe it, so deeply. And not only is it transformed our business, we believe it will transform every single one of yours. And still with that, we really invite you to, take a demo, ask questions, engage us. This is really the time, for AI to really jump into the fray. And I'll be honest in saying, we've been on the we've been taking, small language models, applying them to data, and taking even the models that have been in market and start to edit and work with them. And we've never gotten the performance that we have as the large foundational models have gone through. They are phenomenal. So think about Anthropic, OpenAI, Gemini. These models are phenomenal, and they're getting better and better. And our responsibility is to then think about how we bring that deep, enriched health care context, do the orchestration, build the evals bench, make it so powerful that all of you can make the AI your own. So this is the time to think about it. This is the time to push your organizations around it because there is a huge opportunity to get better, faster, smarter by using Marmot. And so with that, I wanna say thank you. We are gonna do a q and a. Webb and I are on for a little while. We'll start to see some questions coming in. And so, Webb, I'll hand it back to you for any final comments before we launch into to q and a. First off, I just wanted to echo Arif's comments and say thank you all for joining us for your time, interest, and engagement. The, the q and a panel has been on fire, and so it's great to see all these questions. I think one thing that Arif shared, which is really important is for us, we we went through many of those sleepless nights working on those projects, and there was choice, autonomy, creativity, and how we executed on those projects. And one of the things that I'm always blown away by and by the way, one of the questions, in the chat was was along the lines of what personas is Marmot designed for? And and Arif and I both kind of smiled when we saw that because one of the things we talk about at Komodo is the ability for our platform or full stack thesis to democratize access to insights. And so I'll very candidly say, Arif went and learned Python on his own back in the day. I did not. But I worked with so many of you for so many years on so many of these projects that I know what questions I would ask if I was in your shoes. I know generally how I would solve and execute on those projects, And yet at the same time, when I use Marmot, I really feel like a superhero because Marmot will tell me three other things that I didn't think of. It will give me recommendations on, other ways to perform the analysis in a more sophisticated rigorous way, and that's incredibly empowering. And so I just wanted to say super excited to, to have the opportunity to share more with all of you and grateful for your time and interest. Awesome. Well, I think we should probably get a few questions in, and we see a bunch of things that have come through chat. Webb, do you wanna just kick us off with a question and, tell us what and and direct us? Let's go through. We've been we've been getting hit up both, privately as well as publicly here, and so we'll take some of these. Art, did you wanna add anything on that first question we got around the personas that it's designed for? Yeah. Because the go ahead. Yeah. So it so every company is a little bit different on how they wanna think about it. We see some that really want to put the control in the analyst and say, look. The analyst really should be answering the questions and sending regulated information out to the broader audience. And we have some companies that are like, hey. I want my CCO to have access to this. I want my head of marketing to have access to this. We've seen huge variance. And what's beautiful about Marmot is that you choose. You wanna keep this in the hands of the analyst and have them validate every single answer. Absolutely. Marmot will support you on that journey. We the way we provision, the way we create kind of, user tiers and consumer tiers. Like, there's there's ways for us to do that to give you a high amount of control, and there's ways for us to fully democratize access to the entire company for people to just ask it anything without the care or worry that you may have that they give you an answer that our analytics team doesn't feel great about. And that decision is really yours to take. And what's I mean, what what we really stand behind is, like, we will work with whatever framework you want. So we have a lot of clients that are like, look. We wanna empower the entire BI and A team, the Better Business Insights and Analytics team. They are the ones that control this. And we're like, great. We will set this up for you to have that kind of controls. And that's one of the reasons we went on the Maplap, journey is that they want regulated analytics. They wanna make sure that, they are governed well and that there's high trust in them, and Marmot supports that mission. So exactly as Webb said, we wanna make it available for anyone to ask questions. We also respect those boundaries, and we are willing we can configure it in any way that makes sense to to our clients. Actually, Arif, this is analogous, and I think it's something we you can also speak to here. One of the questions was, are the analyses run on a single user account, or is it possible to share your analyses with someone else? It it's all how you wanna do it. So we you can absolutely use, someone to share with. Every chat, every project has the ability to share and make them a user, a viewer, an owner of that project, and that's the easiest way to do it within your company. And so when we set up your account, you'll have the ability to choose who and what you share to. And that's that's been a big part of our our our journey. What's also cool is you can build on someone else's chats. You know, if I'm having a conversation, one of our, the leader of our, our our our CI function, which does all of our work kind of, clinical insights, we had this whole conversation, with Marmot where she was, you know, kind of pressing it on writing this white paper around lung cancer. And I jumped in with a bunch of questions, and she jumped in with a bunch of questions. And we ended up with this amazing paper that we collaborated on together with Marmot. And so you have that ability to, work with someone else's chats. You also have the ability to kinda keep them all very private, and that's what's, what's important about creating the flexibility for enterprise customer. I love that. Alright. Next question. This will also be for you. It's great. Is Marmot based on a single agent or a multi agent framework? And then the follow-up, which isn't directly related is, I'm curious when you say there's an independent reviewer. Oh, it's many it's many different agents, and it's many, many different tools. The hardest part of orchestration is actually getting the agent to use the right tool in the right situation on the right dataset. Think about how complex this product problem is in our world. Like, when do I have relevant lab results or EHR results? When do I use claims? When do I use projected data? And so, the important thing decisions that it's making, first of all, is, like, on the selection of data and how to use the data. But then the reason we employ a multi agent framework is we want an agent that's responsible for that independent review. In fact, it's a totally different LLM because we find that every one of the foundational models is strong at something. Some of them take a lot of data and reduce it down to, like, really good context. Others are really good at challenging the notion of of something. So all of us are using, like, multiple agents, and, and they flow flow into the into the into your workflow in different ways. Like, you can call, their their planner agent to come in and just do a review of where you're at. There's a separate agent that's actually also doing the independent reviews. So it's a multi agent framework with many, many different tools, with really strong orchestration. That's really the way that I would describe the the backbone. Awesome. Let me look down. I I would also say this. We talked about this during the demo, but Marma is delivering transparent, verifiable results. Right? Going back to RF's demo, every output that you're generating can be traced back to its underlying dataset and methodology. And so you all can see exactly where the insights are coming from. And so there's many agents and many tools that are helping build that transparency and trust, but that was at the core of how we wanted to empower all of you. Alright. Let's go down and look. We did that one. Next one, I this is this is interesting, by the way. I I think this is great, Art. Can you run predictive analytics through your prompts? Does it reason? Does it provide context? And does it provide insights from the data or datasets? Yeah. Absolutely. We have a Python tool that's actually, both doing, visualizations as well as building out more complex models. And we're also working on an R tool that's doing the same much more for, our, for a HORWE audience. You will look at the predictive models it chooses, and you will have a lot of feedback on it. And you can also actually upload your own code, and your own models and let it work with your model. And so, that's a really, you know, positive feature of, of of Marmot, which is that whether it generates a model for you or you generate the model and share it, it can run it. And I think that's one of the things that we've had a lot of success in bringing, this to data science audiences because it allows them to have the control over what they wanna do. The other aspect of this is that as the foundation models get better and better at building predictive models, the amount of, fixes that the AI will have to do on its own code start to go down. They it still is an art, and I will argue that they're not, as qualified to do some of the more complex, models, and that's still kind of in your hands. What you gain is you gain a thought partner that's iteratively working on the code and running it procedurally through the data so you can get better and better outcomes. We really like that, as an assistant to you, in model building. Actually, that's this is a perfect segue to the next one. Is it possible to modify the code or make adjustments to parameters such as updating the SQL queries or changing the ICD codes? I mean, this is this is Absolutely. Or You know? Or you can actually, yeah, you can do this with natural language in ways you couldn't before. Like, oh, this is the definition you used for MSCLC. Well, I don't like that. And I'm gonna actually just give you this mutation data, and I want you to use it. So we have, situations where folks are cross token on onto our patient ID, and then you can actually, like, upload a dataset that's patient ID compatible and certified, and, like, you're able to do those comparisons, with very light prompting of the dataset. What the LLMs need is just some context on the data. Like, hey. This is, you know, mutation data derived from these types of, these types of genomics, and then this is the way that it samples. This is the way you line them up, and it probably will take you twenty minutes to construct an LLM on on prompting that additional dataset, and then it runs it. And we've done this with a number of our partners already that are giving us great genomic or EHR data or lab data. And that's really one of the most powerful aspects of Marmot, which is that you can bring in any other third party data so it's impromptu, and then you can apply your own rules, your own code definitions, your own business rules. You can upload them, and the system will execute against it. So we really think that's, designed for flexibility as well as opinionation. And as you kind of build your own methods and approaches in there, they become yours. They're really not there's we don't you don't have to default to any Komodo method that you don't want to, and I think that's what's beautiful about the AI. Alright. We are running up on time. I'm gonna I'm gonna share two thoughts before we close out. The first one is we were inundated with questions. There's over 50 questions from the group here, private and public, that we are it's gonna be impossible for us to get to all of them in the next three minutes. Please, or if I don't know if we still have the slide with contact info, but, obviously, if there's additional info, hit the get demo button. Someone from our team will reach out to you one on one. There you go. Your next advantage is here. Someone from our team will reach out to you one on one, and we'll be able to share a lot more in addition to, chatting with all of you regarding your exact interest and how you want to potentially, embrace Marmot in your enterprise. And the last question is probably my favorite one because, I'm just like that. But the last question that we did field was, I'm sorry if I missed this, but how can I get access to Marmot? And so I feel like that is the perfect question for us to close out on. Art, Obviously, we we both thank all of you for your, engagement over the course of the past hour. Anything else you wanna share, Arith? No. I I think, just more broadly, this is an incredible time for AI and health care analytics. And, whether you use Komodo every day or you were a customer of ours years ago, we'd encourage you to think about, how AI can change the way that you do health care analytics and, would urge you to check out Marmot. It's been, such a labor of love for us, but as well, we see the results with our customers, and having, AI that's built for the health care analyst as their, as their partner. And, you know, we we can't wait to to show you more, but thank you for your time and attention today. Thanks, everybody. Take care. Have a great rest of the day. Take care, everyone.