 From the CUBE studios in Palo Alto in Boston, connecting with thought leaders all around the world, this is a CUBE conversation. Hi, I'm Stu Miniman and welcome to this special CUBE conversation. I always love when we get to talk to founders of companies when they're drilling into some interesting technologies. Want to welcome a new guest to the CUBE as well as one of our CUBE alumni sitting right next to me on the screen. First of all, we have Martín Casado who is a general partner within Driessen Orwitz. Martín, great to see you. It's great to be here. And you've brought along Mike Delbalso who is the co-founder and CEO of Pecton, recently out of stealth, going to dig into a lot of the ML discussion. Mike, thanks for joining us. Thanks for having me on. All right, so Martín, look, you're no stranger to being a founder yourself. We've loved having you on the CUBE over the years. I have to get, since we're getting you on here in 2020, we of course need to start with the fact that there's a global pandemic going on. And I'm curious from your standpoint, from an investment standpoint and looking at technology, how does this make it a little bit different in 2020, say, than you would have thought coming into the year? Yeah, so I think there's kind of a near-term answer and a long-term answer. I think the near-term answer is people don't really understand what the broad impact is going to be. And so companies in the portfolio and the guidance that we do is to be conservative with cash. Let's see how Q2 plays out. And then let's figure out the right way to kind of operate the company in light of the macro changes. Long-term, however, it's very clear that every digital transformation project right now is being fast-tracked. And as a result, we think it's a huge boon to infrastructure as who's been for the software, right? Where in the past, you could deal with kind of legacy setups that were on-prem. This is just not the case anymore. So take for a company like Tecton, Mike's company, like there's a lot of conversations that happen now where the company's like, wow, we really need to have our infrastructure digitized and it all needs to be in the cloud, it all needs to be remote and so forth. So we're actually seeing a ton of tailwinds even though there's uncertainty on the macro environment in the near future. Yeah, you make some great points, Martin, absolutely. The companies that have actually gone through some digital transformation, the goals of that is number one, I should be data-driven. Number two, I should be able to be much more agile and that's what we need in uncertain times is to be able to react fast and answer it. Mike, unfortunately, I've talked to plenty of companies, you can't necessarily choose when's the right time to launch a company, when right time to do an IPO, trying to time the market, but sorry to say, interesting times are upon us. So let's talk a bit about Tecton, give us a little bit of your background, the team, the core team, I believe, coming out of Uber with the Michelangelo project that led to Tecton. Yeah, great, so at Tecton we really focus on what we call operational machine learning, which is really about helping organizations really use machine learning in an applied context, really powering customer experiences, powering business processes, things that really make it to production. And so we help these machine learning AI efforts get past the finish line. And a little bit about the background of me, I used to work at Google, I Google as a product manager for the machine learning teams that power the ads auction. So the models that choose which ads to show and run in real time and are highly productionized and are really core to the business. And then I was at Uber after that and Uber helped start the first centralized machine learning team. And it was really the whole journey for Uber, going from just starting to getting to tens of thousands of models in production. A big component of that was a lot of the technology that we built there, the platforms and infrastructure that we built to support the different business teams to be able to embed machine learning into their product. And so what we're talking about all these very applied use cases, real time fraud detection, ETA estimation, the search pricing, all these things that you think about with Uber. So through that journey of supporting and helping them get to 100 with machine learning, we built out this platform called Michelangelo, which is really a machine learning platform. It's really an end to end machine learning platform. And learned a lot of lessons as we helped out dozens of teams go through the whole life cycle starting a project, pay, what does this mean? What is my business problem? How does it translate to a machine learning problem all the way to having a model in production, monitored and fully, like really fully productionized and kind of growing core to that business. So we learned a lot of lessons from building that tech ton. My co-founders are the other leaders of that project. And we learned a lot of really important lessons that lead to the success of these machine learning projects. And we're now focused on helping a lot of other organizations really start up their machine learning efforts and get these things into production. Yeah, Martin, maybe you could give us a little bit of context here. When I think about repeatability of solutions, how much they scale, there's only so many Googles and Ubers out there. And when I look back at the big data world, there wasn't a lot of repeatability. It seemed like everything was custom, so what did you see with tech ton? What are you looking at in the ML space that made them such an attractive investment? Sure, so maybe let's just pull back and talk about what's going on in systems and infrastructure in general. And I actually think this is probably the biggest shift, certainly I've seen in my career, which is it used to be if you looked at a system, let's say it's Uber, but whatever system, the correctness of that system and the performance of that system and the compliance of that system and the security was dictated by the code that you wrote, right? You wrote bad code, you made bugs, you had vulnerabilities in your code, that would dictate the system. But more and more, that's actually not the case. I mean, these days kind of performance, accuracy, security compliance is actually dictated by the data that you feed into these, right? And so like, you create these models, you feed the data models, the data gives you output and the data that you feed in and like your workflow around those models are really dictating things like pricing or things like fraud, these really important things. And unlike code, we don't have the tools to manage data in the same way. And so if you think of it, we're moving kind of from this code economy to this data economy, data more and more dictates the correctness of all of these systems. And we're talking about trillions of dollars of more data, but if you actually look at the tooling around it, it still feels like the 70s around code, which is like, you've got few thumbs and you've got a lot of tribal knowledge. And so we've been tracking this trend for a long time. We're investors in Databricks, as you know, we've got a large data portfolio. I mean, it's very obvious if you look at what's happening with the cloud data warehouses, if you think like, redshift, BigQuery and stuff, like, the world is going data and pulling extracting information out of data. And so on the backdrop of that, we're like, okay, so you need to be able to think of data like you think of code and have the tooling around it that help makes the lives of people working with this stuff simpler, especially for the core use cases, which is M-Holl and AI. And to that end, I think that this is probably known in the industry, but like looking in the leading companies is like a crystal ball into the future, right? Because they tackle a lot of the problems before the rest of the industry did. And Michelangelo was very well known as the leading project in this. I mean, it had a broad set of respect from the community. It kind of created this notion of a feature store which has now been replicated. And so, really, this is like the preeminent project in one of the biggest macro transformations. And then beyond that, we met the team, they're fantastic. We've got great chemistry. We've got a lot of similar backgrounds. And so the investment was pretty straightforward from that. But I do think it's important to frame it in the context of this macro shift that's going on. Yeah, it's been, can't be overstated how important data is. I do think we need a new analogy probably with what happened with the global pandemic. Everybody was talking about data being the new oil and oil's pretty deep right now. And data is definitely not losing its value. Mike, when I read some of the discussion about, you know, Teccon enables data scientists turn raw data into production ready features and predictive signals, you know, it sounds really impressive. So help us understand kind of the core thing that you do and, you know, where we are in the product life cycle. Great, well, a machine learning application, there's fundamentally two components to it, right? There's a model that you have to build that's going to make the decisions given a certain set of inputs. And then there's the features which end up being those inputs that that model uses to make the decision. And common machine learning infrastructure stacks really are split into two layers. There's a model management layer and a feature management layer. And that's an emerging pattern and some of the more sophisticated machine learning stacks that are out there. And what we built at Michelangelo, we really had this model management layer, this feature management layer. And we recognized the feature management layer was the thing that really allowed us to go from, you know, not just zero to one, but one to one to N and scale out a machine learning across a number of different use cases and allow individual data scientists to own more than just one model in production. And so really what's at the core of that is a few components. The first is just feature pipelines. So these are data pipelines that plug into the business's raw data via batch streaming, real-time data and turn those into features that are these predictive signals that models consume. The second part of that is a feature store which catalogs these feature transformations, catalogs these pipelines and stores the output raw data or the output feature data. And then the third component is feature service making those features accessible to a data scientist when they're building their models and to the models in the production environments when they can make these decisions sometimes needed in milliseconds for real-time decisioning that is quite common in a lot of high-value machine learning applications. So what TechCom really is is a data platform for machine learning that manages all the feature data and feature transformations that allow an organization to share the predictive signals, these features across use cases and really catalog these and understand what they are. And secondly, get these into production so they don't get hung up in that final stage right before they're trying to cross the finish line with the machine learning project. All right, and Mike, the product today, my understanding, private beta, you do have some customers at that point. What tell us a little bit about that? Yeah, we're in private beta with a number of customers. We just went into full production with Atlassium last month. A couple of other customers that I maybe shouldn't name on the air, but we are spending time engaging in kind of like deep hands-on engagements with different teams who are really trying to set up their machine learning on the cloud, figure out how to get their machine learning in production and it tends to be teams that are trying to really use machine learning for operational use cases. We're really trying to drive real business decisions and power their product, customer experiences, and not as much, a lot of the kind of like research, algorithm research stuff, but we're really trying to solve these core data problems that help or that are preventing machine learning projects from being successful. Yeah, it was interesting. Martin, as I was listening to some of what Mike was saying, I'm like, okay, it's not quite the analogy of micro-segmentation or separating the control plane or the network plane and networking, but there were some analogies there. What I want to ask you though is the role of data. I talking to Andy Jassy a couple of years ago, I asked him, the flywheel for AWS for years was customer, how many customers they could get and I was wondering, does data become that new flywheel and there's the center of gravities and the customers that can tap in and monetize what's going there. So I'm just curious your thoughts on that. So I think people don't appreciate how different data is than code. And so I just want to start there because I think it's really germane to this topic. So listen, code is like a finite stage, right? It's like, it's lines of code, you can build it, you can modularize it. It's like building a house. And so the tools that you put around code kind of reign in what's already a fairly low entropy system, like a fairly orderly system. Data on the other hand, data is like the natural world. It's all of the complexity of the universe, right? It's the behavior of humans. It's temperature readings, it's like, you know? And there is so much more complexity and there's so much more entropy in data that the way that you deal with it is so fundamentally different than you ever do code. And so we've had all of these. And so I just wanted to start that with is we've had all of these analogies about data is the new oil, data is where there's the value, et cetera, et cetera, et cetera. But a lot of it's tautological, meaning yes, of course there's value in data. Yeah, yes, if you have propriety access to data, you've got propriety access to data. But what we don't really know is how do you take data and rein it in so you can use it in the same way that you use software system? We actually don't even know how to do that. And so talking about things like data network effects and extracting data is a little bit preliminary because we still actually don't even understand like how much work it takes to mine insights from data. What I do know you need, I do know you need the tools to do it and I do know that those tools are quite different. And so I think that we're now in this era building the tooling that is required to extract the inside of that data. And I think that's a very necessary step and this is where tech time comes in to provide that tooling. And I think once we have a better handle on that, then we can start asking these deeper questions which I think are great questions but the things like how defensible is data? Do you have network effects with data? Can you put in a finite amount of effort and extract signal at all times? Like how messy is data, et cetera. And so I think that's kind of where we are in this journey which is exactly why you need companies like Tecton to help answer this. All right, so Mike, there's been the promise of really unlocking data now. It's been a really interesting discussion point for the last five or 10 years. The company is named Tecton when I've read some of the blog posts and talk about the Cambrian explosion and changes there. So give us, if we're looking forward, you've just come out of stealth. What is success for Tecton two to three years out from now? Yeah, I think the biggest thing is we're trying to help organizations recognize their data. It really is an asset and treat their features like assets. And when we can get to a point where organizations, that teams that want to use machine learning and production don't need to throw a million data engineers at a problem, and we can get to a point where machine learning is not really getting to a special team of experts that are super expensive, that you kind of leave in the corner of your building and hopefully come back 18 months later with some project that is showing some value, that would be success for us. We really are dead focused on the problems that are preventing these projects just from getting into production. And so when we see the industry as a whole have seen success with these machine learning projects, I think we will have our mission accomplished. All right, Martino, I'll give you the final word as to the opportunity you see in front of Tecton. I honestly think the data industry is going to be 10x the compute industry. I just think like with compute, you're building houses from the ground up and there's a ton of value there. I think with data is you're extracting insight and value from the universe, right? It's like the natural system. And every company has data and lots of data and all of it has some information. And so I think that this is a chance to be a very, very pivotal company in democratizing access to data. So I think that the opportunity is enormous. Well, Martino, thank you for joining us again on the update, Mike, thank you. Welcome to being a CUBE alum. Definitely hope to have you back soon to track the journey, congrats on step one out the door and best of luck going forward. Thank you. That's great. Thanks, Stu. All right, be sure to check out thecube.net for the upcoming events that we have today. They're all virtual, but the interviews are all there as well as all the archive. I'm Stu Miniman and thank you for watching theCUBE.