 Welcome back to theCUBE's coverage here in Washington DC for AWS Public Sector Summit. I'm John Furrier, host of theCUBE's an in-person event, but of course we have remote guests. It's a hybrid event as well. Amazon is streaming, Amazon Web Services is streaming all the key, some of the keynotes. Of course, all theCUBE interviews are free and streaming out there as well on all theCUBE channels and all the social coordinates. Our next guest is Melissa Pavlik, President and CEO of PavCon joining me here to talk about predictive maintenance, bringing that to life with the U.S. Air Force. Melissa, thanks for coming in remotely on our virtual CUBE here at the physical event. Excellent, thank you, good morning. It's a great show, people have been face to face for the first time since 2019. A lot of people remote, calling in, checking things out. Kind of an interesting time, right? We're living in, so what's your take on all this? Sure is, I mean it's a new way of doing business, right? I will say I guess for us as a company we always have been remote. So it's not too much of a change, but it is definitely challenging, especially trying to engage with such a large user community such as the United States Air Force who isn't always used to working as remotely. So it's definitely a unique challenge for sure. Well, let's get into it. I love this topic. You had a real success story there as a case study with the U.S. Air Force. What's the relationship? Take us through what you guys are doing together. Sure, absolutely. So we started working with the Air Force now about five years ago on the subject and predictive maintenance. Sometimes you might hear me catch myself and say CBM Plus, condition-based maintenance. They're synonymous, they mean the exact same thing basically. But about five years ago, the Air Force was contemplating how do we get into a space of getting ahead of unscheduled maintenance events? If the military, their big push always is readiness. How do we improve readiness? So to do that, it was a big ask of, do we have the data to get ahead of failures? So we started on this journey about five years ago, as I said. And frankly, we started under the radar. We weren't sure if it was going to work. So we started with two platforms. And of course, when I think a lot of people hear predictive maintenance, they immediately think of sensor data. And sensors are wonderful data, but unfortunately, especially in an entity such as the Air Force, not all assets are censored. So it also opened up a whole other avenue of how do we use the data that they have today to be able to generate and get ahead of failures? So it did start a really great partnership working not only with the individual, I'll say Air Force entities at Air Force Site Cycle Management Center, but we also worked across all the major commands, the individual units, supply, control, logistics and everyone else. So it's been a really great team effort to bring together all of those, what typically would be rather segregated operations together. Yeah, they've been getting a lot of props lately on a lot of their projects across the board. In this one in particular, how did you guys specifically help them modernize and get this predictive maintenance thing off the ground? Absolutely. I think quite simply it was what really we put their existing data to work. We really wanted to get in there and think about they already had a ton of data. There wasn't a need to generate more. We're talking about petabytes of information. So how do we use that and put it into a focus of getting ahead of failure? We established basically three key performance parameters right from the beginning. It was we knew we wanted to increase availability, which was going to directly improve readiness. We needed to make sure we were reducing mission aborts and we wanted to get ahead of any kind of maintenance costs. So for us, it was really how do we leverage and embrace machine learning and AI paired with just big data analytics and how do we take, frankly, what is more of a World War II era architecture and turn it into something that is in the information age? So our modernization really started with how do we take their existing data and turn it into something that is useful and then simultaneously, how do we educate the workforce and helping them understand what truly machine learning and AI offer? Because I think sometimes there's, everyone has their own opinions of what that means, but when you put it into action and you need to make sure that it's something that they can take action on, right? It's not just pushing a dot and moving numbers around. It's really thinking about and considering how their operations are done and then infusing that with the data on the backend. It's awesome. It's, you know, the old workflows in the cloud. This is legit. I mean, physical assets, all kinds of things and there's legacy is also, but you want the modernization. I was going to ask you about the machine learning and AI component. You brought that up. What specifically are you leveraging there on the AI side of the machine learning side? Absolutely. For us, most and foremost, we're talking about responsible AI in this case because unfortunately, a lot of the data in the Air Force is human entry. So it's manual, which basically means it's open and right for a lot of error into that data. So we're really focused heavily on the data integrity. We're really focused on utilizing different types of machine learning because I think on the surface, the general opinion is there's a lot of data here. So it would open itself naturally into a lot of potential machine learning techniques. But the reality of the situation is this data is not human understandable unless you were a prior maintainer, frankly. It's a lot of codes. There's not a whole lot of common taxonomy. So what we've done is we've looked at both supervised and unsupervised models. We've taken a whole different approach to infusing it with truly what I would say, arguably is the most important key element, domain expertise. Someone who actually understands what this data means, so that way we can end up with actionable output, something that the Air Force can actually put into use, see the results coming out of it. And thus far, it's been great. I mean, Air Mobility Command has come back and said we've been able to reduce their my caps, which are parts waiting for maintenance by 18%. That's huge in this space. Yeah, it's interesting about unsupervised and supervised machine learning. That's a big distinction, because you mentioned there's a lot of human error going on. That's a big part. Can you explain a little bit more, because was that to solve the human error part or was it the mix and match because you had the different data sets? But why both machine learning modes? So really, it was to address both items. Frankly, when we started down this path, we weren't sure we were gonna find. We went in with some hypotheses and some of those ended up being true and others were not. So it was a way for us to quickly adjust as we needed to again put the data into actionable use and make sure we were responsibly doing that. So for us, a lot of it, because it's human understanding and human error that goes into this, natural language processing is a really big area in this space. So for us, adjusting between and trying different techniques is really where we were able to discover and find out what was gonna be the most effective and useful in this particular space as well as cost effective. Because for us, there's also this resistance. You have to have resist the urge to want to monitor everything. In this case, we're talking about really focusing on those top drivers and depending on the type of data that we had, depending on the users that we knew were going to get involved with it, as well as I would say the historical information, it really would help us dictate unsupervised versus supervised. And going the unsupervised route, in some cases there's just still not ready for that because the data is just so manual. Once we get to a point where there is more automation and more automated data collection, unsupervised will definitely no doubt become more valuable. Right now though, in order to get those actionable, that supervised modeling was really what we've found to be the most valuable. Yeah, and that makes total sense. You got a lot of headroom to grow into with unsupervised, which is actually harder as you know. Not everyone knows that, but I mean, that's really the reality. So congratulations. I got to ask you on the AWS side, what part do they play in all this? Obviously the cloud, they have a relationship with the Air Force as well. What's their role in this particular maintenance solution? Sure, absolutely. And I'll just say, I mean, we're really proud to be a partner network with them. And so when we started this, there was no cloud. So today, a lot of opportunity are things we hear about in the Air Force where like cloud one, platform one, those weren't in existence five years ago or so. So for us, when we started down this path and we had to identify very quickly a format and a host location that would allow us not only handle large amounts of data, but do all of the deep analysis we needed to, AWS GovCloud is where that came in. Plus, it also is awesome that they were already approved at IL-5 to be able to host that. We in collaboration with them host a NIST 800-171 environment. And so it's really allowed us to grow and deliver this impact out to over 6,000 users today on the Air Force side. So for us with AWS, it's been a great partnership. They obviously have some really great native services that are inside their cloud, as well as the pairing and easy collaboration among not only licensed products, but also all that free and open source that's out there. Because again, arguably, that's the best community to pull from because they're constantly evolving and working in the space. But AWS has been a really great partner for us. And of course, we have some of our very favorite services I'm happy to talk about, but they've been really great to work with. So what's the top services? So for us, a lot of top services are like EC2s, workspaces, of course S3 and Glacier are right up there, but you really enjoy working across glue, Athena. We're really big on, we find a commercial service we're looking for that's not yet available in GovCloud. And we pull in our AWS partners and ask, hey, when's this going to get into the GovCloud space? And they move pretty quickly to get those in there. So recent ones are definitely Athena and glue. Well, congratulations, great solution. I love this application because it highlights power of the cloud. What's the future in store for the AWS Air Force when it comes to predictive maintenance? Sure, I mean, I think at this point, they are just going to continue adding additional top driver analyses through our work for the past couple of years. We've identified a lot of operational and functional opportunities for them. So there's going to be some definitely foundational changing coming along, some enabling new technologies to get that data integrity more automated as well so that there isn't such a heavy lift on the downstream when we're talking about data cleansing. But I think as far as predictive maintenance goes, we're definitely going to see more and more improvements across the readiness level, getting rid of and eliminating that unscheduled event that drives a lot of the readiness concerns that are out there. And we're also hoping to see a lot more improvement and I would say enhancement across the supply chain because we know that's also an area that really they could get ahead on. You know, part of our other work is we developed a five year long range supply forecast and it's really been opening some eyes to see how they can better plan, not only on the maintenance side, but also supporting maintenance for more logistics and supply. Great stuff, Melissa. Great to have you on, President CEO, Pat Kahn. You're also the business owner. How's things going with the business, the pandemic? We look like we're going to come out of it. Strong that the tailwind with cloud technology, the modernization boom is here in GovCloud, 10 years celebrating GovCloud birthday here at this event. How's business, how's you doing? Good, everything's actually been, we were, I guess, fortunate. As I mentioned at the very beginning, we were remote companies. So fortunately for us, the pandemic did not have that much of an operational hiccup and being that a lot of our clients are in the federal space, we were able to continue working and amazingly we actually grew during the pandemic. We added quite a bit of a personnel to the team so we're looking forward to doing some more predictive maintenance across, not only expanding the Air Force, but the other services as well. Yeah, the people who were agile, had some cloud action going on, were productive and they came out stronger. Melissa, great to have you on theCUBE. Thank you for coming in remotely and joining our face-to-face event from the interwebs. Thank you so much for coming on theCUBE. Great, thank you, John. Have a great rest of your day. Okay, I'm John Furrier here at theCUBE with AWS Public Sector Summit in person and remote, bringing guests in. We got the new capability of bringing remotes in, we do in person, obviously in face-to-face here at theCUBE and it's like to be here at the Public Sector Summit. Thanks for watching. Robert Hushabin.