Montana may seem an unlikely place for cutting-edge sensor systems and computational solutions to the big data problem, but there are a cluster of companies with Alex Philp at the helm, that are on the forefront of that problem space. Matt Ball spoke with Philp about GCS Research, Adelos and TerraEchos, as well as the trends with systems and sensors that are giving rise to a whole new awareness of our interactions. The TerraEchos solution is particularly compelling with its ability to rapidly process large volumes of data to aid decision support in real time.
Ball: I find it inspiring how you’ve made a transition from academic research to an entrepreneurial bent, founding several successful companies. What’s the history of how that came about?
Philp: I entered the University of Montana grad school here in Missoula, Mont. in 1992, and my Masters was in geography with an emphasis on GIS. I started my doctorate in historical landscape ecology using GIS and remote sensing, and halfway through that I had an opportunity to work at the Goddard Space Flight Center on the convergence between remote sensing data types, satellite-derived data products, geographic information systems, as it was becoming early web enabled, and early Internet browsers as a portal to the World Wide Web. We built a great team of people that were active between 1998 and 2002.
As part of that four-year process, my doctorate essentially shifted into telling geographic stories in a meaningful way to support business process or educational experience. I tied that all around the Lewis and Clark Bicentennial and developed that map story with systems within systems within systems that were doing much of what is called the cloud now. We had geospatial distributed analytics, distributed systems, map services, and terabytes of data spread across machines to educate and tell that story. That was all fairly applied GIS research.
I started GCS Research in 2002, and very early on I was able to bring in outstanding business partners. Mike Beltz and John Waterman, two GIS people from the University of Montana, became partners in GCS Research, and we built that company into what it is today. A few of the interesting things that came out of that is that we built the first Google Street View before Google really knew what it would be, and before they acquired Keyhole. The history there is to be bleeding edge, and be leaders in geospatial technology solutions.
Ball: When did you branch into sensors?
Philp: I started to think about sensors everywhere after working with sensors on satellites and aircraft. It wasn’t new to me that we’d have an explosion in mobile GPS. I became very interested in sensors, and sensored systems, and elements of the sensor web. I had an opportunity to work with the Navy on a project with a very high-end fiber-optic sensor. Think of it as thousands of microphones buried in the ground, using fiber optic cable to support those digital sensors. We licensed that technology from the government and formed a new company called TerraEchos to focus on that activity in 2006. By 2008 we had our first prototype, and then through 2011, we were building and delivering systems to the U.S. government, working with the Navy and the Department of Energy.
We were doing next-generation physical security. The idea being that you need to create a physical perimeter using five or six different types of sensors -- cameras, microphones, geophones, magnetometers, biometric arrays, millimeter wave radar, airborne assets. As our work evolved, I realized that it wasn’t enough to focus on just our sensor system, but that we needed to interoperate and interact in real time, both spatially and temporally, in identifying, classifying, tracking and localizing that threat. We started extending and expanding that into Adelos, which is the Greek word for hidden. We proved some phenomenal things with that system, and gained some recognition for it.
Ball: Getting recognized by IBM is certainly a validation of a high order in terms of your technical expertise. How did the alliance with IBM come about?
Philp: I became aware of an IBM technology that they referred to as “System S,” and if you recall that their original code name for the first relational database management system (RDBMS) was “System R,” then it’s pretty significant. The technology is now known as InfoSphere Streams, which is a paradigm shift in how we do computing. Streams is a compute platform that allows you to do massive continuos query of data in motion, and so you’re dealing now with the perfect technology to analyze structured or unstructured data, data that is either continuous or discrete, and it can deal with any data from any source. It’s a shift in how we think about associative analytics.
I’m interested in associative memory and analytics that leads to synthetic solutions, theoretically, conceptually or otherwise. You have multiple sources, you do interdisciplinary calculations, you do computationally-intensive work in real-time, and what comes out are answers that provide automated and semi-automated predictions and decision support. All of that we were doing around Adelos, and we achieved some major success in 2010 and 2011, to the extent where we earned significant industry awards from IBM and others based on the innovation.
To give you an idea, we were running a system where we were doing 785,000 Fast Fourier Transforms (FFT) per second, we were consuming more than 350 MB per second from over a thousand sensor channels, and we were conducting detailed classification analytics on that data to tell you what something is, and to do that in a 14th of a second. Twelve thousand times a second we were achieving these data matrices and doing these statistics. A year ago that was taking 27 minutes, and now it’s taking a 12th of a second. When we were done with our work, we analyzed more than 20 petabytes of data per day.
Most people don’t understand it, and don’t believe me, including elements of the government that have a hard time believing that we did this. This is groundbreaking stuff.
Ball: How have you capitalized on all the recognition?
Philp: We were interested in going somewhere with this analytical capability, so venture capital investors took a serious look and we created a whole new company just around the central nervous system of Adelos. We literally ripped the cerebral cortex out, and named that product Kairos. Kairos is the Greek word for right timing. The Greeks had chronos and kairos, with chronos referring to sequential time. Kairos is a convergence of different events occurring in time that must come together in order for a catalytic event or moment of insight to occur. If you miss that right time, those factors may never come together again, which is a perfect word to describe what we do.
We take data from multiple sources -- some of it is big, some is slow, some is small, some is structured, some is unstructured, some is video, some is text, some is audio, etc. -- and we bring all that together into an associative and correlative processing engine. We run computationally-intense analytics, including Markov analysis and Bayesian analysis, to make sense of all this data.
We now have three companies that I’m involved in. The TerraEchos name is now associated with the new company that was launched in January of this year that focuses on big data in motion analytics, with a specialty in spatiotemporal real-time analytics. The company that had done all this work has become Adelos, with a focus on advanced sensor systems. Then there is GCS Research, that focuses on geospatial analytics. I spend my time between the three, with big ideas and big ambitions, based on 15 years of work that we’ve been doing with an interdisciplinary approach.
Ball: Is geospatial an input to the Kairos big data in motion product?
Philp: It is and it isn’t. What we’re trying to figure out right now is where your standard geographic information system begin and end, and where does Kairos begin and end. As you know, your traditional geographic information systems with classic database-driven GIS, is not very good at dealing with these kind of compute problems. I’ve been working for years on trying to inject sensor systems into GIS, and to greater and less success. We’ve been doing that over and over again for customers.
The best thing that I can say is that we pull data from GIS technology, primarily Esri, whether it is a file, a table or a database as a source, and we write and publish back to the source. We’re pulling what’s meaningful and necessary from GIS as part of our calculation, and then when we have an answer we put the answer back into GIS to support those workflows and systems. We’re not trying to make a GIS analytics fusion machine, we’re trying to find the right interoperability at a technical level with reading and writing the right raster, vector and record as part of a service. We’ll be showing and talking about that at the upcoming Esri User Conference in San Diego.
Ball: How does the Kairos product fit in terms of the IT ecosystem, between systems?
Philp: We’re trying to position our capability in the right part of the technology workflow or technology stack so that we handle certain parts of the information problem, but we’re not trying to be everything in terms of an IT solution. We’re trying to focus on that part of the problem that is occurring now, and be able to consume ungodly amounts of data and then at the end of the calculation you have an answer.
Theoretically, we’re leveraging the database in terms of our knowledge and understanding of the past and our records, and we’re using that along with our models and current information, to do a better job of predicting and forecasting. That’s where we’re spending a lot of our time right now as a startup, to leverage the building blocks that we’ve developed, and to put that together into a new solution set that we call the Kairos technology platform.
Ball: I’m fascinated by the application of this technology to existing business problems, but also making sense of details that you may never have had an insight into. There seem to be so many opportunities, how do you go about focusing on specific industries or problems?
Philp: I’ll be the first to admit that I’m really happy and pleased that we have other business executives that are helping us do a better job on productizing and bundling this capacity and capability that we’ve invented to address the right markets in the right time in the right way.
There are several megatrends there as it relates to computer science. The ubiquitous computing devices first revolutionized our ability to produce information back with mainframes and that has continued on today with more and more mobile devices. We next addressed how to compress, store and treat information. Then the Internet provided the means to share information and communicate with our computers. We started producing, then we started organizing, and now we’re figuring out how to share.
What we’re seeing now as a megatrend is what do we do with all this, to squeeze insight and knowledge from the data. The Internet of Things is upon us, and everything is becoming embedded with a sensor. I have nine sensors on my Android phone right now that is producing an incredible amount of information that is waiting to be harnessed and utilized for a variety of applications, and there are 900,000 new Android devices being activated every day. The amount of opportunity right now in terms of the convergence between device, data, and analytics is incredible.
To make sense, and extract meaning today, requires the right combination of software, hardware, analytics and network. I describe the opportunity as the 3VI over network equation, that is the convergence of high volume, high velocity, high variety (3V) with the exponent of intelligence over network. This is the kind of equation that we’re creating with Kairos. What we’re working to identify are specific industries that are hitting the wall on velocity, variety, volume, analytics or network, and we’re trying to bring our solutions to them to achieve a hundred to a thousand to even ten thousand-fold improvements on optimization.
Primarily it’s about saving time. If my problem currently takes me 34 seconds, and if I can derive that value in one one-thousandth of a second now we’re really starting to buy time. We’re starting to manipulate time for existing and past functions, and do a better job of reacting. But, we don’t just want to react better, we want to be more proactive for our customers.