This transcription is provided by artificial intelligence. We believe in technology but understand that even the smartest robots can sometimes get speech recognition wrong.

Solution Showcase: Exploring Private AI Live at HIMSS Tim McNamee and Mark Larochelle

[00:00:00]

I am Bill Russell, creator of this Week Health, where our mission is to transform healthcare, one connection at a time. Welcome to today's solutions showcase on the keynote channel where we spotlight innovations, making real impact in health systems. Let's take a look at what's working today.

Bill Russell: Alright, here we are from the HIMSS floor and we have a solution showcase. I'm looking forward to this conversation. Um, I'm joined today by, uh, Tim and Mark with, uh, Cloudera. I, I'm gonna have you guys introduce yourselves and, and your role and, uh, and then I really wanna get into what CLA Cloudera is doing.

Um, for me, this is an update, like you guys are gonna update me on where you've come.

Tim McNamee: Yep.

Bill Russell: Um, I know your, your data background and I want to hear where you guys are going on the, a AI side. Before we get there, please introduce yourself and then what do you guys do?

Tim McNamee: Yeah, so I'm Tim Magni. I'm the account manager for both the BA and HHS.

So I get the pleasure of working both on the provider, payer and life sciences side of the house [00:01:00] here for Cloudera. Okay.

Bill Russell: F

Tim McNamee: answer.

Mark Larochelle: Uh, I'm Mark La Rochelle. I'm a state, local and higher ed and healthcare specialists working with healthcare providers, uh, in the US including UPMC and and MGB.

Bill Russell: Fantastic. Uh, well, as I said, so Cloudera has a, a very interesting backstory.

Everybody knows you as data player, cloud player, and whatnot. Um, walk us through the, the AI journey, or just the journey over the last couple years. I'd like to be up to speed and I think some people listening would like to get speak.

Tim McNamee: Yeah, so Cloudera was the original kind of big data company. Our founders came out of, uh, Google, Facebook, and Yahoo, who were kind of challenged with managing massive amounts of data.

So the founders came out and, you know. Wanted to bring the technology to the commercial marketplace. We were quickly brought into the CIA incubator program to help with the intelligence community, but, uh, both financial and life sciences, uh, as well as healthcare our, our largest verticals today.

Bill Russell: I was introduced to Cloudera back when I was a, uh, CIO of St. Joe's. And we were using your platform through [00:02:00] one of our partners, but the entire backend was built on Cloudera. And I didn't know that at the time. I, I went back to him later, I'm saying, Hey, what, what was the architecture?

He goes, we, we just, all that, all the really cool stuff. And we were doing AI at the time. We were doing some machine learning stuff on top of it. On top of your existing platform. Uh, you know, give us an idea of how your approach to AI differs as a, as, uh, organization with a data background as opposed to maybe a, a.

A pure play AI company.

Tim McNamee: Yeah. A lot of the pure play ais are focused on their own models and Right. Perfecting those models to help bring value to organizations. Where Cloudera, what we're focused more on is kind of the ability to use any open or closed source model. And so these are important for things like deep learning, orent ai, um, and giving you more of a variety.

You know, our, our other big focus, and this is kind of from our roots, is around governance and lineage. Right. It's making sure that you can track everything that the AI is doing. Um, agencies are extremely concerned about rolling [00:03:00] AI out and, you know, doing that ethically. And so with Cloudera, we kind of enable that.

Bill Russell: I see. I've been here for 24 hours. I've heard the word observability transparency about 50 times, and it's, it's one of the big things that healthcare, and I'm sure, I'm sure the government's looking at that and your payer clients are looking at that as well. That's a very, I important. Uh, aspect of what they're looking for, especially, uh, transparency from a healthcare perspective.

They wanna know how, how the data's being used, how it's coming up with, uh, some of the decisions and whatnot. I wanna talk about a use case for fraud detection.

Tim McNamee: Yep.

Bill Russell: how big of a problem is this?

Tim McNamee: Yeah, it's a big problem. They, they estimate it's about a $300 billion problem here in the us and really they can, you know.

Auditors can really only look at about 1 to 2% of the, you know, payments that they're making. So it's a, it's a huge problem

Bill Russell: from a, from a healthcare standpoint, you know, we had all sorts of, uh, mechanisms, rule rules, agents and whatnot looking at these things. Uh, a lot of things got through the, [00:04:00] cracks there.

I mean, how, are you guys approaching this?

Tim McNamee: Yeah, so the rules, uh, was really a flawed model. The fraudsters could get past the rules. And really those, those full models were pay and then chase, right? And we all know that once those payments are made out, then the fraud is already out and that you're never gonna get the money back.

So, um, with Agentic AI it's more reasoning. So what we're trying to do is take that initial claim, compare it against the medical records, um, looking for duplicate payments, looking for any types of anomalies that would flag that, and then quickly alerting a team of people. To audit it before the payment is made.

So it's just a, a huge shift.

Bill Russell: Is that, uh, primarily Medicare or Medicaid, or is it, is it commercial? It's, it's just across the board.

Tim McNamee: It's across the board. I mean, both, uh, providers are, are concerned about, you know, the problem and certainly the payers. Uh, but Medicare is, is, you know, quickly trying to address it.

They have been coming out with some new philosophies and methodologies. There's a new wiser model [00:05:00] that they're using that would fit well. Into an eent uh, workflow as well. So, you know, and if you look at the current administration, I mean, it's, it's a big focus. We we're seeing, you know, Medicare, medicaid problems across state and local, where the payments are, it's a very fragmented data problem.

So, you know, this is where eent AI can really create specialized agents to really kind of look in some of these disparate silos of data.

Bill Russell: What are the clients of the state and local, what are they, what are they focused in on from a, from an AI perspective or a fraud prevention standpoint?

Mark Larochelle: Yeah. It, it, it, it's very similar from the fraud prevention perspective. I think a lot of those customers are also still trying to get their data together. So they wanna apply ai, what you've gotta have your data, you gotta have that, uh, governance throughout so that you can get more from your AI agents and output and know that it's verifiable.

Trace your sources. And those sort of things. So I think a lot of state and local customers are still in that getting the data AI ready stage and others are much more advanced and they've got all their [00:06:00] agencies sharing data. So you can, you can do more to prevent fraud if you're taking data out of the silos and you have all the agency data together in one warehouse, or the ability to bring your AI to those asylums.

Bill Russell: Yeah. It, it's, it's interesting. When I came into healthcare, that was the problem we were trying to solve, and I'm still talking to people. That's the problem we're still trying to solve.

Mark Larochelle: Yeah, yeah.

Bill Russell: There's, there's still a ton of silos, uh, that exist within the environment. Uh, Uh, talk a little bit about the agent approach.

I mean, what does the agent do when it finds something and how does it, how does it respond to those things? How does it find it? I mean, how is, how's the whole thing work?

Tim McNamee: Yeah, it, I mean, it, it's really in. Important, you know, from a n agent perspective, that you do have the human in the loop, right?

You can't trust AI to make decisions on massive payments or on a, you know, a clinical outcome, right? So, you know, we run it through the models. We quickly produce, you know, some type of actionable result, um, for either an auditor to take a look at and make a, a go, no go decision on the payment. Maybe dive deeper [00:07:00] into, you know, what, what was seen.

Or on the clinical side, having a doctor, you know, agree with the opinion that the agent AI comes up with. You can't, you know, we all know you can't a hundred percent trust ai. It needs to have it human in the loop. And, you know, I think that's an important element that most healthcare organizations are focused on.

Bill Russell: So talk to me about explainability. Why is it important and what does it, what does it mean in this process? ,

Tim McNamee: It, it's everything. You can't say, oh, AI said that it was fraud. Right? Um, and that's, you know, the biggest challenge, right? So we. Through our data lineage and our data governance, we're able to prove this, you know, prove that the action happened, show the evidence, and give you a hundred percent explainability so that you can take it to the court of law.

Right. And, and justify why you did not make a payment. And, and it's extremely important.

Bill Russell: Yeah. You see that defense happening, Hey, the AI made me do it, kinda of thing that, that would be, uh, that'd be interesting. Uh, the, the agent layer concept, I mean, is that. Uh, Cloudera concept or is that an industry, uh, movement that we're [00:08:00] seeing?

Tim McNamee: You know, it's, it's an industry movement for sure. I think the approach that Cloudera is taking though is, you know, allowing you to use any models within your Egen workflow. So if you find that Lama is gonna do better, or clawed is gonna do better in implemented into certain phases of the Egen. 'cause Egen is doing a lot of different things.

Um. You know, Cloudera kind of differentiates themselves from, from being really an open architecture, but also really focusing on that governance and observability that we've talked about. Because again, you can't just let your AI run robes. You need to have, you know, the explainability of a robe.

Bill Russell: I think some people are thinking, well, well, why don't we just use the latest version of, of Claude, or the latest version of Open ai?

But it turns out some of the other models are better at, uh, you know, at categorizing things or you know, just, and they're very fast. Inexpensive and they're able to do those things. Having that ability to move between models

Tim McNamee: Yep.

Bill Russell: Is, is so key. Uh, I would think, um, you know, the other thing that's interesting, and one of the reasons that we went with Cloudera 10 years [00:09:00] ago, um, was we didn't like the idea of our data being strewn all over the, all over the world and all over the country.

Talk a little bit about, um, and, you know, the, the, the, the ability to set up a, a private data network, if you will, on Cloudera. Yeah,

Mark Larochelle: yeah. We're leaders in private ai. We can bring the AI to the data, whether it's in your data center or your cloud tenant. So that's different than pushing a copy of your data to a, a, let's say a hosted vendor from a security perspective, uh, as well as a control perspective.

And then that governance, we also extend that to the agents. So you've got agents triggering workflows with other agents, so having some, some type of, of manageability, observability, and governance. On the agents and the AI models themselves gives the healthcare customers a a lot more control and, and a lot better security.

Tim McNamee: Yeah. And a lot of these healthcare organizations have a, a lot of facilities. Right. I think you ran a, a, maybe a 13, you know, uh, different facility. But one of the challenge challenges they have [00:10:00] at the VA is, is that it's a, it's a, is

Bill Russell: it, is it more than 13?

Tim McNamee: It's 174 and they have a lot of latency issues. So with Cloudera and our inference services that we offer, we're actually able to host the model close to the data and bring the model to the data.

So giving you that performance, um, and, and privacy that you, you might want in healthcare by bringing the model to the data, whether it's on-prem, on the edge. Or even in the cloud, but say within your VPC,

Bill Russell: One of the questions that keeps coming up in our meetings is, what, what does, what does governance, what does strong AI governance look like?

Mark Larochelle: I think what's unique to Cloudera's governance capability is, is to have it from everywhere, from ingest, from, whether it's, uh, instruments, bedside patient, whatever the, you know, so from end to head.

Exactly all the way through, and then including the AI model management. I think that's, that's what we have is that common lineage governance and security throughout. And that's much different than sort of the integration [00:11:00] tax when you're trying to track the governance across a lot of different platforms.

So I think that gives our customers an advantage and it helps get, you know, it's just that one at UPMC with their digital health summit and there's so many interesting researchers that have these specialty AI models. But to get them out there at scale and the clinicians using them, right, and having central it manage them, I think that's where we, we can help accelerate that and accelerate the trust the clinicians will have.

Bill Russell: So what, what do you think, uh, we will close on this. I mean, what, what do you think the people listening to this, what would surprise them about Cloudera or, uh, what you guys are doing today?

Tim McNamee: Yeah, I, I mean, with a lot of our CIOs, it's kind of like an aha moment. You know? They're like, I just, I had no idea that I could.

Post private AI within our enterprise. And really what we're seeing is that the more models grow and the more AI you're using across your organization, it gets expensive in the cloud. With hosting these models, with Cloudera, with Cloudera ai, um, [00:12:00] not only we give you that governance, but we give you the scalability and the performance and it gets sustainable cost, right?

So particularly in public sector, we have a challenge of, you know. If the money's not there, you gotta just stop doing it. And so with predictability of cost, I think is something and, and our ability to kind of bring AI to the data where it resides and protect that PHI data, if you will for sure.

Bill Russell: That's fantastic.

Mark Larochelle: Yeah. I just met with the CTO with a large provider and that was, you know, their understanding. They were a little bit surprised even as a 10 year Cloudera customer that you can take these private AI models that might normally require you to push Gator to their host gig. Cloud and you can still use those private AI models against your data, either in your data center or in your tenant.

And that's a big, uh, cost saver, not having copies of data, uh, along with a, a better security model.

Bill Russell: That's fantastic. Well, I wanna thank you guys for your time and love the, uh, uh, work you're doing and look forward to hearing more as, uh, as time goes. Five.

Tim McNamee: Yeah. [00:13:00] Thank you. Thank

Bill Russell: you. Think

Mark Larochelle: can build free soon.

Speaker 10: thanks for watching this solution Showcase on Keynote with me, bill Russell. We believe every healthcare leader needs a community they can lean on and learn from. Discover more solutions and join our community at this week. health.com/subscribe. Share this with someone who could benefit from these insights.

Thanks for listening. That's all for now.