>> From the Aria resort in Las Vegas. It's the Cube! (upbeat music) Covering AWS Marketplace. Brought to you by Amazon Web Services. Hey, welcome back everybody. Jeff Frick here with the Cube. We're at AWS Reinvent 2018. I don't know how many people are here, 60 000, 70 000, your guess is as good as mine. I'm sure we'll get an official number shortly. We're kicking things of here. Three days of coverage. Monday, Tuesday, Wednesday, Thursday, that's four days. We're at the AWS Marketplace and Service Catalog Experience here at the Aria. We're excited to be kicking stuff of with Rajeev Dutt. He is making AI that makes AI. We're going to get into it. He is the CEO president and co-founder of DimensionalMechanics. Rajeev, great to see you. >> It's great to meet ya. >> How many Reinvents have you been to? >> This would actually be my second. >> You're second? >> My second. Yeah, it's like- it's- I always feel really energized after coming here. It's like- last year was like heavy AI centered. >> Right, right. >> It was just really all these sessions in AI was really exciting. >> Let's get in to it for the folks that aren't familiar with DimensionalMechanics. What are you guys all about? >> So DimensionalMechanics is about lowering the bar for entry like to most people. So that's kind of our first focus. Our second focus is to make sure that deployment strategies allow you to deploy across any end device. So it's basically intended to be a complete end-to-end capability. >> Around AI? >> Around AI. >> The Artificial Intelligence. >> The Artificial Intelligence. >> Yeah, yeah, yeah. >> Most important part. >> Okay. >> Yeah, so, it's about reducing the bar for entry for Artificial Intelligence so that anybody without even a machine learning background can build very sophisticated models on our platform. In sometimes as little as 14 lines of code. It's just incredibly easy. We've had high school students use us, we've had university professors, who have nothing to do with AI, use us without any problems. And, really the way we do that is that we have an AI that we call the Oracle. We are all Matrix fans. (Jeff laughs) And so what this- the Oracle does is it has a vast knowledge base, has a lot of additional machine learning components and things like that. That essentially allow it adapt and learn based on the kind of problem you're trying to solve. So, every time it solves the same problem, it gets better and better at what it's doing. >> So, so, um... Is it, is it libraries, is it pre-configured, are there specific type of application that it works better on? What's kind of your go to market? >> So basically, think about AI studio as a full server application. So it, what you essentially do- we created our own language called the NeoPulse Modeling language. And the NeoPulse Modeling language, think about it as sort of the SQL for Artificial Intelligence. It does a lot of very complicated things in just a couple of lines. So essentially what you do is you compile it on the machine so when you write the NML code, the NeoPulse Modeling language code. You compile it on the machine, it looks at your data which is sitting in a bucket. It starts training the model. Once the model is ready, you can export the model as a PIM object, so Portable Inference Model object which is one of our creations. And that allows you then to deploy it on to any end target as long as it's running on runtime. And on runtime can be basically sitting in the cloud or on a device. Sometimes we're also looking at right down to FPGA kind of device levels as well. So, extremely low power devices as well as cloud computing, but gives you that flexibility, but it also, which is really important, it makes AI accessible. So anybody without like any background in it- My wife is a radiologist and she's actually looking at using it for her own internal usage... >> And how much do you have to learn? You have to learn the NeoPulse language, right? >> The NML language is really easy to learn. So we had a high school student who spent about a week learning it and so a week later she was ready to start coding and she has built her first models using that. And the way it does that is that you actually, we have a keyword auto inside NML which is context aware, and so when the compiler sees auto it goes out to the Oracle and says hey, I've seen- this person needs help building an architecture or figuring out what function to use or what hyperparameters to use and so on and so on. And the Oracle will come back and say hey, use this architecture, use these hyperparameters, use these settings or functions or these optimizations in your model and... >> So is that doin' that when I'm setting up the model in the first place to give me directions or is looking at the model once I've spun it a couple times and saying wait, this looks like one of these, maybe you should do some of this. >> So what it will look at is your data. So it will actually look in to your data, the type of data, how much data you have, the kind of problem you're trying to solve, how many, for example, if it's a classification problem, how many classes you have, and all of that basically determines the kind of model that it will use. You can also specify the level of complexity that you're interested in, like, are you interested in a very simple model, a complex model, is over fitting a risk at all It will determine all these things behind the scenes >> Right, right. >> based on the kind of problem that you're trying to solve. And the first time it solves it, it will give you a pretty good answer. It's usually very good, but then the second time you solve it or a third time you solve it, it gets better and better and better, because it's able to learn from its mistakes. So, and eventually it gets really good at its job so. >> But it's still, but it's a still a model that I built for that application. You're not drawing kind of pre-configured models down from the Oracle. >> No no, you're basically training it from scratch. >> Right. >> It's entirely intended for custom models. So companies that are- have highly customized data, like radiology or for example, looking at wind stress patterns like in polarized light and stuff like that. So things that are not normally covered by the standard image recognition and so, using things like transfer learning or fine tuning doesn't help in this particular case because if you've trained a model in dogs and cats then like, training it to recognize stress patterns, is just not- >> It's not going to work. >> It's not going to work. >> So you got to prepare for your interviews, looking through your website. You list a really dramatic example of where using your guys technology was like, I don't know, a tenth of the price >> Yes, yeah. >> And I think one month versus six. >> Yeah. >> I wonder if you could share some couple examples that, you know, people are putting this to use. >> Okay, so, we have actually a few. So one of them is with a company. They're focused on kind of a resume matching, so we built them- they were initially quoted by another company at around 450 000 and they were warned that they would not be able to exceed 40% accuracy given the data that they had. We managed to get to about 83/84% accuracy for about under 10 000. So that was like a huge huge reduction. Then the second one was just recently, another company had been spending quite a bit of time and resources on building out a technology to measure heart rate. We were able to look at that and produce, instead of spending like their 20 000 a month or so, we could bring it down to 4000 in total. So these are the kind of sort dramatic reductions in cost that our platform can offer. Stanford University, another great example. These are physicians that we're working with. None of them have any engineering background like, for them, Linux is in itself- That was the hardest thing for them to do was to get used to Linux and so once they start building on our platform it was like they actually built a model that was good enough that they were able to publish at the RSNA, which is like one of the biggest radiology conferences in the world. In this case it was for Pet CT, which is a three-dimensional model because there's a three-dimensional image if you will >> Okay. >> of the human body and so was able to determine whether somebody had a tumor or not and I think they mananged to get, with a very limited data set, about 74/75% accuracy and this was actually at Stanford, so it's a pretty, pretty big name. >> Right. So, Rajeev we're here at AWS Marketplace Experience. You're still a relatively small company. I think you said you had a good size C round, gettin' ready to go out and get a decent A round. >> Right. >> What does it mean to work with a company like Amazon? I mean, as a small company, just to get, just to get an approved vendor set up at Stanford, probably not an easy thing, right. There's all kinds of legal Ts and Cs. >> Exactly. >> As a startup their always worried about whether you're going to be around tomorrow. >> Exactly. >> So your part doin' AWS, so how's that been workin' with AWS and the Marketplace . >> Well firstly, it's definitely given us the Amazon backing in a way, so when people see you're on AWS, they see that connected to you, that automatically gives them a little bit more confidence. >> They vetted you so you must be good. >> Exactly, exactly. And the second is that it gives access to a market that we otherwise wouldn't have had like, if I'm thinking about like producing software that you have to download on our website, that's a very very limited market. You have to attract people to your website and so on and so on. Now it's like we're on the Amazon- there's a machine learning hub on AWS. We're on that, so which means that when people search for machine learning, our name does come up. >> Right, right >> It means it's very easy to launch. You don't have to worry about setting up a machine, worrying about how to configure it. Everything is done automatically, makes life really easy. >> Right. >> On top of that, the AWS team has been- the Marketplace team has been really extremely helpful connecting us with end customers. So very often they will refer people to us. In fact, one of our largest customers came through an AWS referral, so for us it's been nothing but a win-win. >> Right. What about the potential downside? Not to rain on the parade but the old joke used to be if you're a start-up makin' widgets, you know, you just got your first order with Walmart the good news. Bad news is you just got your first order with Walmart. That's opening up a huge global distribution opportunity, I mean in theory, you know, say you got a 1000 customers tomorrow, that might be a little bit of a challenge. >> Yeah, so we actually are starting to hit that. So, we- so our version two was really our go to market version and- which came out earlier this year, and so we've been trying to like wrap up the sales on that side and literally in the last three months. It's like I have not been home for six weeks now because I've been in the far East and traveling and, it' like- because of this heavy customer interaction at this point. So we have a very good story to tell the investors now, like, this has also helped with the investments rounds that we're actually looking at. So we have a very good story to tell the investors that you know, our like invoice list and so on is huge at this point so we need help now. It's actually more about raising, like building up a team now than it is about can we get orders. >> Right. It's really delivering more than sales. >> Exactly. >> I see what you're saying. And so we need to build up a delivery team, we need to- I mean, it's fairly intuitive, but at the same time it's a new technology which means, as with any platform, you're building up a team of evangelist, support individuals and so on. And there's going to be a marketing component as well, so we haven't really driven marketing that much. AWS has been great in kind of doing some of that for us, but we need to of course very actively go out and market. We haven't had that capacity yet. >> All right. We look forward to watching the story unfold and thanks for spending a few minutes with us. >> My pleasure. Thanks, thank you very much. All right, he's Rajeev Dutt, I'm Jeff. Thank you for watching the Cube. We're at the AWS Marketplace and Service Catalog Experience at the Aria, come on by. Thanks for watching. (upbeat music)

(bouncy, melodic music) - The world's kind of revolting against these mega-siloed platforms. - That's the risk of having such centralized control over technology. If you remember in the old days, when Microsoft dominance was rising, all you had to do was target Windows as a virus platform, and you were able to impact thousands of businesses even in the early Internet days, within hours. And it's the same thing happening right now, as a weaponization of these social media platforms, and Google's search engine technology and so forth, is the same side effect now. The centralization, that control, is the problem. One of the reasons I love the Blockstack technology, and Blockchain in general, is the ability to decentralize these things right now. And the most passionate thing I care about nowadays is being driven out of Europe, where they have a lot more maturity in terms of handling these nuisance-- - You mean the check being driven out of Europe. - Their loss, - The loss, okay. - being driven out of Europe and-- - Be specific, we'd like an example. - The major deadline that's coming up in May 25th of 2018 is GDPR, General Data Protection Regulation, where European citizens now, and any company, American or otherwise, catering to European citizens, has to respond to things like the Right To Be Forgotten request. You've got 24 hours as a global corporation with European operations, to respond to European citizens, EU citizens, Right To Be Forgotten request where all the personally identifiable information, the PII, has to be removed and auto-trailed, proving it's been removed, has to be gone from two, three hundred internal systems within 24 hours. And this has teeth by the way. It's not like the 2.7 billion dollar fine that Google just flipped away casually. This has up to 4% of your global profits per incident where you don't meet that requirement. - And so what we're seeing in the case of GDPR is that's an accelerant to adopt Cloud, because we actually isolate the data down into regions and the way we've architected our platform from day one is always been a true multi-tenant SaaS technology platform. And so there's not that worry about data resiliency and where it resides, and how you get access to it, because we've built all that up. And so, when we go through all of our own attestations, whether it's SOC Type One, Type Two, GDPR as an initiative, what we're doin' for HIPAA, what we're doin' for plethora of other things, usually the CSO says, "Oh, I get it, you're way more secure, now help me," because I don't want the folks in development or operations to go amuck, so to speak, I want to be an enabler, not Doctor No. - I'm a developer, I search for data, I'm just searching for data. - That's right. - What's the controls available for making sure that I don't go afoul of GDPR. - So absolutely. So we have phenomenal security capabilities that are built into our product, both from an identification point of view, giving rights and privileges, as well as protecting that data from any third party access. All of this information is going to be compliant with these regulations, beyond GDPR. There's enormous regulations around data that require us to keep our securities levels as high as we go. In fact, we would argue that AWS itself is now typically more secure, more secure, - [Mike] They've done the work. - than your classic data center. - [Mike] Yeah, they've done the work. - AI-ers, explicable machine learning. - Yeah, that's a hot focus, - Indeed. - or concern of enterprises everywhere, especially in a world where governance and tracking and lineage, - Precisely. - GDPR and so forth, so hot. - Yes, you have mentioned all the right things. Now, so given those two things, there's normal web data, NML is not easy, why the partnership between Hortonworks and IBM makes sense? Well, you're looking at the number one, industry leading big data platform, Hortonworks, Then you look at a DSX Local, which I'm proud to say I've been there since the first line of code, and I'm feeling very passionate about the product, is the merge between the two. Ability to integrate them tightly together, gives your data scientists secure access to data, ability to leverage the Spark that runs inside of Hortonworks Glassdoor, ability to actually work in a platform like DSX, that doesn't limit you to just one kind of technology but allows you to work within multiple technologies, Ability to actually work on your, not only Spark-- - You say technologies here, are you referring to frameworks like TensorFlow, and-- - [Piotr] Precisely. - Okay, okay. - Very good, now, that part I'm gonna get into very shortly. So please don't steal my thunder. - So GDPR you see as a big opportunity for Cloud providers, like Azure. Or they bring something to the table, right? - Yeah, they bring different things to the table. You have elements of data where you need the on-premise solution, you need to have control, and you need to have that restriction about where that data sits. And some of the talks here that are going on at the moment, is understanding, again, how critical and how risky is that data? What is it you're keepin' and how high does that come up in our business value it is? So if that's gonna be on your imperma-solution, there may be other data that can get push out into the Cloud, but, I would say, Azure, the AWS Suites and Google, they are really pushing down that security, what you can do, how you protect it, how you can protect that data, and you've got the capabilities of things like LSR or GSR, and having that global reach or that local repositories, for the object storage. So you can start to control by policies. You can write into this country, but you're not allowed to go to this country, and you're not allowed to go to that one, and Cloud does give you that to a certain element, but also then, you have to step back into, maybe the sorts of things that-- - So does that make Cloud Orchestrator more valuable, or has it still got more work to do? Because under what Adam was saying, is that the point and click, is a great way to provision, right?