Gregg Masters 00:05
PopHealth Week is brought to you by Health Innovation Media. Health Innovation Media brings your brand messaging alive by original or value added digitally curated content for omni channel distribution and engagement. Connect with us at www.popupstudio.productions. Welcome everyone. I’m Gregg Masters, Managing Director of Health Innovation Media and the producer and co host of PopHealth Week. Joining me in the virtual studio is my partner, colleague and lead co host of PopHealth Week, Fred Goldstein, President of Accountable Health LLC. On today’s show, our guest is Sri Ambati the CEO and founder of H2O.ai. H2O is a visionary Silicon Valley open source software company that creates and reimagines what is possible, a company of makers that brought to market new platforms and technologies to drive the AI movement. H2O is the leading open source data science and machine learning platform used by nearly half of the Fortune 500 and trusted by over 18,000 organizations, and hundreds of thousands of data scientists around the world. H2O democratizes Big Data Science and makes Hadoop do math for better predictions. Prior to H2O Sri co founded platforma and was the Director of Engineering at datastax. Before that he was partner and performance engineer at Java multicore startup Azul systems, tinkering with the entire ecosystem of enterprise apps at scale. Sri is known for his knack for envisioning killer apps and fast evolving spaces and assembling stellar teams towards productizing. That vision. So Fred, with that introduction over to you help us get to know Sri and his work at H2O.ai.
Fred Goldstein 01:57
Thanks so much, Gregg, and Sri, welcome to PopHealth Week.
Sri Ambati 01:59
Thank you for having us, Fred.
Fred Goldstein 02:00
Yeah, it’s a pleasure always love getting into the areas of AI, healthcare, all the cool stuff. So why don’t we start a little bit, give us a little bit of your background, how you started the company, sort of what the company is doing now?
Sri Ambati 02:11
Well, H2O is at that movement, and we are trying to democratize AI and make AI accessible and bring it to every every person’s fingertips and change lives in the world. democratizing AI. AI has historically been around with us for a very long time, much before we all are here. But I think it was, the haves and have nots of algorithms have been kind of having a big difference. And so our mission is to make every every company in the AI company and make them use it, we use open source to do that, and made it started with an incredible pool of talent. And this was one of the world’s top data scientists, mathematicians, physicists, software engineers, a college to a home. And what it does is essentially make it really simple to both process large and small datasets, and produce insights and decisions from data.
Fred Goldstein 03:10
And you’re in a lot of different industries. Obviously, we’ll focus today on healthcare, but some of the other places you work in or people that use your products?
Sri Ambati 03:18
So we’ve bought a good third of our customer bases, actually in financial services, and insurance. And we found that the one of the top use cases for us is fraud prevention, or subprime credit scoring. democratizing credit is just as passionate, we’re just as passionate about it. But as you can see, health is the most important wealth. And most people lose their credit history or lose their path towards a safe retirement because of improper health. And some of the core themes behind our health has been healthcare has focused on was more on care, we want to focus more on health. And the optimizing function we want to have for this industry is health divided by care, or how little care does one need to be to live a long fruitful life. We all started with a core genesis of the company started the company because there wasn’t enough tools to process large datasets, especially in oncology, my mom was diagnosed with cancer, and the datasets used for lumpectomy versus mastectomy, were very, very small. So you’re understanding why I was diagnosing the problem of why people are not using world class software to address these problems. And we found that open source was not approved or FDA approved. And so what we did was essentially build a world class software out in the open so the world can use it. And if anything, we have seen a fast forward to the pandemic we have seen an epidemic of dashboards but that’s speaks to how widespread technology and data and data analytics has become. But what the epidemic has also done is fully humbled us in terms of how much more is to be done. So we’re super excited to announce H2Oai helped with some of our, and put together an incredible team and some our core customer base, whether it’s the likes of Kaiser or Hospital Corporation of America, or the open source users. So we begin to start, put our team to help with the mission of being there where we are needed.
Fred Goldstein 05:35
That’s fantastic. And you mentioned some of your your clients, you know, HCA and some others. Can you give us some examples of the work you do in that I know, as I’ve looked through your website, and things, there’s a lot of really sort of population health type focused approaches you’re taking to look at various issues. So what are some of the things you’re doing for these various companies in the healthcare space? Or how are they using your product to develop products for themselves?
Sri Ambati 05:56
At the heart of it, we were essentially, we found ourselves trying to help Pfizer, for example, predict flu, California flu predictions a few years ago. And quickly those problems led us to start looking at predictions for COVID, if you will. And during COVID, a lot of our customers were stressed on their supply chain, PPE or oxygen. And a lot of our customers have been trying to assess high risk groups. And if a hospital was close to a senior facility and multi hospital systems, how do how do you kind of predict you might have people coming to the hospital, or nursing staffing, simple issues, such as even operationally operational efficiencies, then one, the next level, we’ll start looking at clinical trials for some of our other customers, especially rare disease, how to kind of kind of match patients with the right kind of regime, a lot of our customers are looking at oxygen saturation levels, like for example, we work with one of the community hospital, in bulk of what we’ve done is essentially help our frontline workers right should have them in health care workers have been in the frontlines of this war. And we are trying to be on site help them to the extent we can with data and analytics in the eye.
Fred Goldstein 07:19
And I know we’ll get to COVID in a little bit and talk some more in depth about some of the unique stuff you’re doing there. But as I looked at it, I mean, you talked about starting with flu season prediction, but you go into fraud detection, cancer detection, imaging, personalized drug matching, it’s really a broad array that people are using your things for, and I tend to think of companies trying to focus in one area, how is it that you’ve been able to create something thats that, ubiquitous?
Sri Ambati 07:44
Well, we took an iconic name H2O, we’re trying to live up to it to some degree, we want AI to be ubiquitous. I mean, what we repackaging it in different ways. But really, we’re talking about math. And mathematics is the language of the universe, and nature. And we’re trying to essentially use it to get some more reasoning behind get some more pattern recognition to what’s happening. As it turns out, a rare condition is one in a million or 1 in ten thousand cases. And that’s typically how we are looking at data, certain algorithms, like random forest or boosting boosting the signal. These techniques are just as similar to solve when you’re solving a fraud prevention problem. But they’re also just as similar. So we took the best in class techniques applied in financial services, oftentimes, we find ourselves borrowing those techniques and applying them in healthcare or insurance, and see spectacular, low hanging fruit in those spaces. Sometimes it’s about data. So finding good patterns in the data, what we call feature engineering. And we have the world’s best platforms to do that. So so that allows customers to apply problem solve problems in a broad, broad scope.
Fred Goldstein 09:06
Yeah, I know, you talked about the math, and I was reading some of the blog posts. And as they got more in depth into the technical aspects of them, I had kind of lost myself a little bit there, which is really fascinating to me to say, you know, you’ve really looked at this from a technalogical approach, a technical approach, how what algorithms you’re using to go and solve these problems. And, you know, it’s interesting, we recently had an individual with Change Healthcare on the show, and I noticed you’re also doing work with them as well. And and they actually talk extremely favorably about that. Can you talk a little about what you’ve done with Change?
Sri Ambati 09:40
Change is actually a writing a data exchange, which sort of in many ways for all the providers, and we found that and we found that at the heart of all of problem solving is data and Changes brought a ton of centralized repository and what they’ve done essentially is used our algorithms to streamline a lot of kind of both conversion of unstructured data to structured data. And then from structured data applying algorithms to kind of pull out simple insights that power their end users. And it’s, it’s actually exciting to see them merge with one of our favorite customers in our Health Group, in Optum, and both of whom have been very one of the earliest adopters of all open source, H2O. And they were, they’ve given us so many good insights on how to apply AI, whether to look at fast growing vectors, whether to look at fraud prevention in insurance side to helping payers to be providers. So I think that the core of, of the, of what we’re seeing with our customers is applying intelligence to solve problems and predict them as opposed to looking so historically, you have BI, which is looking behind AI is looking ahead for patterns and predicting them. And adapting to them beforehand, the rate at which data is changing, we need machine learning to solve machine generated data. And so I think that’s kind of the the historically we will look at queries to understand data. What we’re doing now is machine learning to kind of predict what’s likely to happen in the data.
Fred Goldstein 11:28
And if if somebody wanted to, say begin to introduce AI into their healthcare system, let’s say it was a provider, how would they go about working with you would they would they get a hold of you, and then you have this platform they can put their data into and you help them do that help them understand the AI behind it, and the predictions that they get out of it? Is that how you work with groups.
Sri Ambati 11:50
At the heart of what we’ve done is we’ve essentially made the product so easy to install and be consumed, either from data science, faculties, or data engineers, if they’re doing anything, in the world of data lakes, and the thing the world of snowflake or online, offline. on prem, they are already probably having a library from us, which is doing their core math. And our practice starts when they started looking into start trying to productionize the usage of this, we auto generate a lot of the production inference engines. So you do two phases in the in the day in the life cycle of data, you’re learning from data, you call it training, and then you’re scoring new data that’s passing by, and that’s called influence. So what we do is auto generate inference engines for these data scientists, so that they can go into production much more faster. And so typically, Today, H2O is almost a necessary skill set for most data scientists doing practice. So open source software is running already on their servers, or in their cloud environments. And then when they are trying to optimize automate more of the functions, look for automatic detection of changing data, and retrain the models, or, or all the ancillary functions needed to do the lifecycle of machine learning. That’s where we’ve announced H2Oai hybrid cloud, this cloud environment runs where the data is, data has gravity. And, and so customers ended up partnering with us, we have a stable of incredible some of the world’s top data scientists called grandmasters are at H2O. So our our data center is partnered with a customer to solve the first few problems, create a simple app, simple dashboard that they can take to their business users, and eventually build their own app stores that enhance the life cycle of customer adoption for us.
Fred Goldstein 14:06
And can you talk about these data scientists, you mentioned, you know, these, their rankings and grandmasters and things? Could you explain to our audience a little bit about what that is, as many of us probably I certainly didn’t know of in advance, and I’m sure many of our audience doesn’t know it as well.
Sri Ambati 14:19
So just like chess, data, science has competitions. And so there was a company called kaggle, which started a community along the same time we started our open source communities. And so we found that these communities would be using the the tools and and libraries and, and platforms we’re building and competing on against each other using datasets, both public and company, given datasets, if you will, if someone wins five gold medals, they become a grandmaster. Though, we have 20% of the world’s top grandmasters at H2O. World number one, Guanshuo, World number three. Phillip Fulmer, World number five Mario’s former world number one, number seven. So you have the top several of the top grandmasters. And there are about 200 of them today, and 20 of them called H2O home. These folks are working closely with our customers finding gaps in our product, improving our products, they’re working closely with, with, with product teams to kind of mimic best practices. What this allows us is to crowdsource our product roadmaps, if you will, it also allows us to learn from the community very quickly, at the heart of AI is feedback loops. If we can connect fast feedback loops, we can learn faster and grow faster. And so for us, by bringing the users into the company, the power users into the company, the feedback loops are faster. And that’s been an incredible tool to both improve the products but also improve the company. Strategically, I would say that many projects in the future will never start without a data scientist at the the the team. In H2O, we say we don’t start a product without a kaggle Grandmaster on the team.
Fred Goldstein 16:20
Wow. So and I find that so interesting to bring that much. It’s always cool to see companies that just bring in a wealth of expertise, and see what then they develop and build. And you can see as you’ve covered such a broad spectrum of healthcare issues. I’m wondering too, obviously, I was I was reading through some of your pieces, you talk a lot about it, it’s always a major issue with health care, cleaning the data, you know, adjusting it, where you don’t have data elements and things like that. And one of the issues that’s come up more recently with AI is the problem of bias within the data. You could you talk a little bit about that. And maybe some of the things you might be able to do or people could do to try and ensure they don’t end up in that kind of mistake.
Gregg Masters 16:58
And if you’re just tuning in to pop Health Week, our guest is Sri Ambati the CEO and founder of H2O.ai. H2O is a visionary Silicon Valley open source company that creates and reimagines what is possible?
Sri Ambati 17:15
Yeah, so data has bias and historical historical data has collected what happened, it is capturing experience. And so it’s important to kind of, kind of rebalance it in some sense. Avoid proxy columns, for example, proxy to raise proxy to gender, that would otherwise be driving to the same conclusions. And so, so I think that I think preventing bias starts with or using simpler algorithms, which is kind of a contradiction in some sense, where we are pushing for deep learning with most accurate models. And as it turns out, there are simpler equivalents in for two very sophisticated algorithms. So we have an automatic model distillation function, which allows you to go from a very sophisticated model, it’s difficult to explain to the business user difficult experts to explain to the physician on why he needs to not give this particular drug to a patient to prevent readmission, you need to simply if we need to simplify it. And so what happens is we’ve created equal ends for the very sophisticated algorithm because we have built for the sophisticated garden and the simpler ones that are in the H2O platform. So so that’s one side, the other side is to actually create, tease out adversarial data testing. So when you create a, how do I, if I pump this change drifted this particular change, will that cause the model to quickly bias itself and go back to doing things that was before. So that ability to create that kind of test framework, so that our validation framework allows us to kind of make it much more simpler. But fundamentally, data has bias it’s not it, we do need the bias to kind of understand and build the models in the patterns, but preventing it from making jumping to conclusions very quickly, whether through safety premium safeguards and safety into the safety guardrails into the how the model processes built in that’s kind of where we build auto ml. And it auto ml for us is not done without being able to be explained the model. And that’s kind of where it’s necessary to understand why the model made the decision is made because models have blind spots, just like humans.
Fred Goldstein 19:53
And you know, speaking of bias, we had the issue obviously, with some bias and some AI and healthcare recently was this is what we’re seeing with COVID, where we have this differential impact on individuals, and we’re seeing, obviously, if you built a model and those individuals weren’t in the data set, you’re going to end up with a different prediction than having the broad base model that includes everybody. And you’ve now moved into this area of COVID, and put out quite a bunch of unique features and products and programs that are, can you talk some about what you’ve done with COVID?
Sri Ambati 20:24
Yeah, I think COVID presented an incredible opportunity for us to take these grandmasters and have them work closely with physicians, hospitals, sometimes community hospitals, were not ready to they was we wanted to be there. last thing you want is the world’s top talent, not working on problems of that are presented today. And I think that gave us a wealth of insights. On first things we look, we found that as our open source was being used in China, at the outset of the break outbreak in Wuhan, and we quickly found out that the virus is just one of the most easiest to spread of many have any any other public health disaster we’ve seen before. As a result, we could pretty much predict from even January last year, that what is the first wave look like? Customers came to us they were trying to ask what the for example, the hydrochloroquine problem or can we use any other rheumatoid drugs to prevent going into ICU, how much drug how much of the drug will be needed to start working some of our customers? Genentech is one of our customers. So we started looking at their public plans to do for supply and demand the other problems in the pandemic impacted toilet paper or cleaning supplies. And so we were looking at demand for cleaning supplies in Brazil, Malaysia and US to change distribution centers UPS is a customer of ours. So looking at our routing algorithms for a UPS truck to all the way to watch what should Walgreens have in their stores. We were looking at flu shot predictions in the past, but now we’re looking at almost even regular staples in people made more visits to Walgreens. We looked at Mobile mobility data, and we’re able to find out that De-urbanization was a now going to be a definitive trend mid crisis, we could predict that folks will leave in New York City and then find calling new places homes, not just in Florida, but along the highways to new places, Michigan, had new immigration, you saw the further outbreaks later on. Walmart’s local Walmart’s local retailers were still having different things just stack up. So we’re sort of looking at the whole economic piece as a whole and TPP loans we’re looking at which which business is likely to come back COVID impacted small medium businesses quite a bit and impacted women more from a work standpoint, a lot of women dropped out, as you can see. But above all, I think it impacted the black and Hispanic community quite a bit in the US. And so we couldn’t use talking about bias we couldn’t use has COVID or has been infected by COVID as a variable to give loans because it was really a proxy for race in many places. And long story short, we found ourselves in the throes of all our customers asking for more AI more analytics, continuously learning systems, not just periodic quarterly models now became weekly models or daily models. And being able to react to those changing conditions became almost a necessary skill set for our customers. And they were hospitals or retailers FMCGs. So that meant that we rebuilt our entire software stack in a way that people can quickly deliver Business Insights by using data and modeling became almost automatic, semi automatic. So that’s what we release released Wave which allows people to build quickly, these dashboards and applications. But we also release our AI cloud as a way to reuse all these use cases into there. We also build an NLP engine that looks at all the publications that are happening on COVID and summarize them on a daily basis. Or summarize then. So researchers can quickly look at what were the real trends on social networks where Twitter was a huge force during this phase. So we started looking at some of those feeds to pull up, what were the topical top of the trending news so that we could start making them available. Physicians found themselves trying to adapt regime on the fly. So we started looking at some of the community hospitals to see how we can take the best, the latest regime that was working better to apply for most aggravated cases versus less more oxygen saturation issues, what could probably predict what is the likelihood of someone running into ICU or not ventilator demand for something that we looked at some of our customers globally, this was a very important, I would say, a significant event of an incredible scale. I think it’s gotten us to reflect more on how we can be more prepared for the next pandemic. And the variation the mutation of the of the of the virus, b 118. We started working with some other epidemiologists, like Nick and others, who basically have been working on both sides of the pond, Imperial College London, and here were the data in other countries has been much more easy to access. Our practice with Singapore or Asia, and China, the data sets were far more easy to pull up and start looking for trends I think the US owes itself to, to transform the public health data systems. And I think, a lot of good changes coming telehealth efficiency of telehealth. That’s the other trend that we’re seeing. And that’s all said and done, I think the event horizon for innovation has been fast forwarded by at least half a decade, if not a decade, because of COVID.
Fred Goldstein 27:16
Well, I think, I mean, there is so much to unpack in the stuff you’ve done, it is just unbelievable. I think what we have to do is get you back on at some point to dig a little bit deeper into some of these individual areas and get some information, some more information about what you’re doing. And that because it’s so broad, it’s unbelievable to listen, you talk different countries, different from this supply side to the demand side to the clinical side, you really dive into a lot of it with the AI tools. So I want to thank you for coming on pop Health Week. It’s been fantastic having you as a guest.
Sri Ambati 27:47
Thank you for having me.
Fred Goldstein 27:48
Back to you, Greg,
Gregg Masters 27:49
and thank you for that is the last word on today’s broadcast. I want to thank Sri Ambati CEO and founder of H2O.ai for his time and insights today. For more information on Sri’s work at H2O.ai go to www.H2O.ai or follow them on Twitter via @H2Oai And finally, if you enjoyed this episode of PopHealth Week, please like us and do consider subscribing to the show on your preferred podcast platform for PopHealth partner Fred Goldstein. This is Gregg Masters saying Bye Now.