Tapan Patel - CTO at Trextel

(Joel Beasley at 00:00:00) Hello, my friends. Today we are talking to Tapan, the CTO at Trextel, and we discuss Trextel's mission to make networks more efficient, Starlink's early release, and using micro pitches to move your projects forward. All of this right here, right now on the Modern CTO Podcast. Here we go. This is the Modern CTO Podcast.

(Joel Beasley at 00:00:31) Hello. Hello.

(Tapan (Deep Patel) at 00:00:32) Hey, Joel. How are you?

(Joel Beasley at 00:00:34) Fantastic. How are you, my friend?

(Tapan (Deep Patel) at 00:00:36) Doing good. I love the beard. I don't think I've ever really watched an episode, so I don't know if this is new. Probably is new because I've seen your photos and you don't look like that.

(Joel Beasley at 00:00:50) I know. That's the big move. We just did some new photos about a week or two ago, and now we have to update all social media sites. I was pretty surprised when the weather changed, so it just got cold here. And you're in Atlanta, right?

(Thapan (Deep Patel) at 00:01:04) Correct.

(Joel Beasley at 00:01:05) I was just up there last, or two weekends ago, and I have family in the Madison area. So we took the kids up to see grandma and then go to the aquarium and do all of that. But when the weather is changing, this being the first year I've ever had a beard, you would think that it would be more warm, right? You think that the beard would provide more warmth. It doesn't as much as you would think. But I saw that you do a lot of running and stuff outside. Do you notice the beard at all with that?

(Tapan (Deep Patel) at 00:01:31) Not really. I think it's a fallacy. It's one of those things that say, oh, you look at old Viking warriors and they have this big thick beard because it's cold. I don't think that's what it was. Even though I've never really grown it that long, but I don't think that's how it works.

(Joel Beasley at 00:01:50) Have you ever met Roy? He's the CTO of Calendly. He's out in Atlanta, and he's a big runner too.

(Tapan (Deep Patel) at 00:01:58) No, I've unfortunately never met Roy. I know Tope, though, and we have talked. So I know of Roy, but we have never actually met.

(Tapan (Deep Patel) at 00:02:08) So Atlanta is a pretty big community running-wise. There's lots of events. There's lots of even local charity events that you can participate in even in this cold. But because of COVID this year, it's mostly been remote. People have tried doing Strava challenges and things like that. It's mostly been quiet this year, though.

(Joel Beasley at 00:02:29) Yeah. When I was talking with him, I run every morning, right? But I do a two to three mile run just to wake up, and that's how I start my day. And he gets on and I'm like, oh, he likes to run. That's cool. And the dude runs 50 miles at a time and competes in the Georgia Death Race. And I was like, okay, well, I don't want to go running with him.

(Tapan (Deep Patel) at 00:02:51) Yeah, that's a different kind of running. I think I'm more in your court. I do five Ks or a five mile run just to kind of know where I stand. And, you know, I have some PRs that I try to beat from time to time if I'm feeling particularly adventurous. But generally speaking, I'm not doing 50 miles at a time. That is quite out there.

(Joel Beasley at 00:03:17) For me, I didn't do the outside stuff or at least make an effort until I got older. When I was younger, I was mostly inside on the computer, and that's where I really got into technology. Where did you get into it?

(Tapan (Deep Patel) at 00:03:33) So I grew up in India. I moved here right after high school. And in India, back when I was, I would say, in middle school roughly, the computers started to become a thing. And so my first operating system was MS-DOS, and, you know, you'd run some funky commands, and our teacher would have coded up a script that we would run, and it would do something like a fun game, things like that. That was my first exposure.

(Tapan (Deep Patel) at 00:03:58) As I grew older in high school and whatnot, it just seemed a little interesting to me. I wasn't one of those whiz kids that was like, oh, I started the computer club or anything like that. But we learned Pascal back then, which I didn't really like. There was another programming language that was specifically made for education and was for younger kids, but they taught us in high school for some reason. It's called Logo. I think someone at IBM made it to teach their kids, and it became a thing. I don't recall who exactly made it. But the idea goes, there's a pixel pointer, and you can do some Lisp-like functional programming to move it around, and it'll draw funny things. So you can say, put the pen down, change the color, now make a stroke, move 10 pixels forward, and things like that. So that's when I really got interested in making funny drawings with programming. So it was quite fun.

(Tapan (Deep Patel) at 00:04:59) And yeah, that's kind of how the love affair started. We played games too, of course. I remember my dad back home took out a loan to buy me a computer because they were very expensive back then. And I was the only kid in the neighborhood that had a computer at home. Other folks would go to cyber cafes and things like that. And so we'd always have kids over at our house playing video games, and that's kind of how the love affair began.

(Joel Beasley at 00:05:28) And so you grew up mainly in India, and then when did you come? Did you come directly to United States? Do you spend time in other places?

(Tapan (Deep Patel) at 00:05:35) Yeah. No, I came directly here. I was about 17 or 18, so right when I graduated high school that summer. So I came here in May, and I think I graduated high school in March. So I got my certification, and then I just moved here. So funny enough, I wanted to be in aviation. That was my big dream. I wanted to be a pilot. So I applied to get into Indian Air Force, and they wouldn't have me. I got into the American Air Force, and my parents wouldn't let me go. And so it's just funny how I came into computers. It seemed at the time that it was a second choice to me, but I moved to America to get into aviation. And I wanted to get my—I had my student pilot license in India, and I was going to collect some more hours to have my private pilot license done here in the States.

(Tapan (Deep Patel) at 00:06:34) And it just didn't quite happen that way, and I ended up doing my bachelor's in computer science and kind of fell in love even more so than when I was in high school. It was more of a hobby back then, not a passion. And really in college is when it started to really become a thing for me. Yeah, I came here after high school. Long answer.

(Joel Beasley at 00:06:53) Are you still flying today?

(Tapan (Deep Patel) at 00:06:55) No, not really. I wish. So our CEO, Jason Morrow, he's a bit of a flyer. He has flown in private planes, and I think some bug got into him, and he bought a single prop plane, which he's asked me many times that we should go on a trip with. That thing is a bit of a death trap because if you've ever flown in a smaller plane—so yeah, once you get up there, it's kind of—it's not the same thing as flying an airliner, right? It just gets up there and slowly you're moving in the air, not really doing much. And, you know, if a gust of wind comes around, it shakes the whole thing. A lot of rattling noises, which can be a bit scary. I, at some point, when I get back into flying, once I feel I have—well, who knows when it's going to be a good time, but just things are very, very chaotic at the moment. And so waiting for things to calm down before I get back into flying.

(Joel Beasley at 00:08:00) Yeah. When you can buy a bunch of planes is when you get into flying.

(Tapan (Deep Patel) at 00:08:04) So, surprisingly, it's not that expensive to get into flying. It's, I think, $140 an hour, something like that. I don't know the current rates this year.

(Joel Beasley at 00:08:13) I looked at it a couple months ago. It's $5,000 to get your basic license and then $10,000 to get a more advanced one. It's not horrible.

(Tapan (Deep Patel) at 00:08:21) Yeah. Yeah. It's not that expensive. It's just a time commitment because you have to study radio telephony, meteorology, and things like that. And it's just been so long—instrument reading—and I just have forgotten probably most of it all.

(Joel Beasley at 00:08:35) Yeah. It's amazing how our memory works too. When I saw that you had founded multiple companies and now you're working at a company that I don't think you founded, Trextel. Correct?

(Tapan (Deep Patel) at 00:08:44) Correct. No, I did not.

(Joel Beasley at 00:08:46) So I really like people that have, or I like getting to talk with people that have that experience similar to me of, you know, starting something from nothing and you just have to deal with everything. It's a very maturing experience to go through. And so whenever I see people then go back and work with other companies, I'm always curious. Is it because they found other entrepreneurs and they wanted to join them? Because maybe they didn't want to start from scratch all over again, or they saw a really hot opportunity in the market. I'm always curious. What was your entrepreneurial perspective to join Trextel and not just start something else again?

(Tapan (Deep Patel) at 00:09:25) Yeah. No, that's a good question. I've read this, or this may even be a cliché, that founders are a little bit broken. And once you're a founder, you're kind of always a founder, right? Something has to be crazy with you to want to go out and start a company when in theory, you know, you can move to Bay Area and make $500,000 a year and have a cushy, comfortable life. So something deep within you has to drive you to either want to start a company or join something small enough where you can really grow. Some people do it for money and all sort of wrong reasons, which—so one of the mistakes that I had made earlier in my career is I started a company, not for the money, but I knew I wanted to start a company, and that was the reason. I didn't care about the problem. I didn't care about, you know, who I'm starting with, and so I kind of made these naive mistakes. And as I've kind of gained a bit more experience, you know, I'm only looking to work on problems that really excite me even if I wasn't getting paid. That's the threshold that has to meet. And one of those problems that Trextel is trying to solve is, if you think about networks, you know, it's more important now than ever. And that can be the copper cable coming into your house or wireless, you know, AT&T signal. It takes a lot to operate these things, and this kind of work happens behind the scenes and is obscure. Most people don't know how it all functions. We just expect it to have very high bandwidth, low latency available all the time. So who does all this?

(Tapan (Deep Patel) at 00:11:03) And I knew one of the founders of the company, Chris Damico, from—we worked together at another startup. And he was on the board of Trextel at the point, but he and Jason Morrow started this company together. And so he was the one who introduced me to the problem. And, frankly speaking, I had never thought about this, right? Who—how much it takes to keep these systems up and running. Even though I'm a CTO, and I run many complex systems before, but our purview is generally within the application layer. Now think about that for much, much grander transcontinental networks, right? It just takes a lot of human labor to keep this up and running. And so this niche was very interesting to me. These support centers are called NOCs, network operation centers. And currently, it's run in a very traditional way where you get an automated ticket by a system or an alert or some customer calls and says, I have this problem. They go in, they look at the different tools, they look at different logs, and say, okay, alright. I think here's your problem. I need you to do X, Y, Z, Mr. Customer. I'm going to press a button, and it's going to fix your issue. We should be able to automate that to a large degree. And, you know, majority of those resolutions tend to be, hey, just reboot the thing. Sometimes it gets complicated. Let's say you're a Walmart and in a store, you may have hundreds of devices, multiple different connections, multiple WANs coming in, maybe a wireless connection, maybe a satellite connection with something like Starlink. How do you manage all that can get pretty gnarly. So I found it really interesting that this company who's been around for 13 years, Trextel has been around for 13 years, and has solved this problem by deploying humans.

(Tapan (Deep Patel) at 00:12:53) And now starting to ask this question, why do we—why can't we make this process more efficient and help others do the same? So, you know, they started building systems to kind of dock food and make our own NOC more efficient. And as we are progressing, there is a natural extension of this to other providers that can be applied. And that's just an interesting problem to me. So removal of repetitive human labor, whether that's physical—Atlanta has another bunch of startups that are attempting in the similar space. And so this idea of eliminating repetitive human labor kind of got into my head, and that's how I got involved.

(Joel Beasley at 00:13:38) That's interesting. Yeah. When you're describing that, it sounds like a kid in the candy shop for an entrepreneur. We have an existing business that we already have validated as a model that's operating, that's highly human intensive, and we want to bring someone in to help us make efficiencies everywhere and grow.

(Tapan (Deep Patel) at 00:13:56) Yeah. And I think, so MSPs as these are called, managed service providers. When you're Starbucks and you buy Internet, you know, let's say you have hundreds of locations, you're not going to go to Comcast here and AT&T here and some other company there. You generally go to one big vendor and say, okay, you know, give me Internet, AT&T. And they have to then figure out how to do aggregation of all these providers, billing, and all of that. So these deals get very, very complicated. And again, it's one of those businesses that's kind of in such a niche market that no one thinks about. So another company, I don't know if you know about this company called Flexport. They do freight forwarding. It's a Valley-based company. There are many, many companies like these. I think they were a YC company as well. They do freight forwarding. Who the heck knows what freight forwarding is? There's a company in Atlanta called Stord. They do kind of warehouse management, but I'm totally describing it incorrectly.

(Joel Beasley at 00:14:58) No, I've had them on the show before. A year or two ago. Yeah.

(Tapan (Deep Patel) at 00:15:02) Not Sean. His cofounder you had on the show, I think.

(Joel Beasley at 00:15:06) Jacob.

(Tapan (Deep Patel) at 00:15:07) Jacob. Yes. There you go. So yeah. So this is one of those industries that can benefit from this type of automation and modernization of software stack, which is also kind of an interesting industry for me.

(Joel Beasley at 00:15:23) So are your customers the AT&Ts who are trying to organize this or who are your customers?

(Thapane (Deep Patel) at 00:15:29) Yeah. So AT&T, who's one of our biggest customers. The word customer gets a little bit tricky here because we build the systems and the software that gets used by AT&T, and it gets used by the end customer. So the example that I gave, Walmart, who's not a customer, but similar retailers like them can buy our software and buy the services that come along with it. And they generally tend to buy it from the likes of AT&T because that's where they want to get one bill with all the line items included instead of having to go and purvey.

(Thapane (Deep Patel) at 00:16:04) So it's a distribution channel for us, but they internally end up using our systems as well. So one of the reasons why it just works that way is because our network's interconnected. If there's an issue and we can't resolve it, let's say it's on the AT&T's end, then we need to be integrated into their systems and know, hey, this is a circuit that's having trouble. So we're going to log this ticket over to you for you guys to work on.

(Thapane (Deep Patel) at 00:16:29) Or, you know, we need to get smart enough to say, hey, this circuit is having exactly this issue. Here's the resolution. If you apply it, it'll take care of the problem. So AT&T is a distributor, and they are customer as well because it's kind of interlinked.

(Thapane (Deep Patel) at 00:16:47) And then, of course, others in retail and banking space tend to be our customers. Security-focused customers that require uptime is ultra sensitive—those tend to be our customers.

(Joel Beasley at 00:17:01) And what's the buzzword in this area? I was reading some of the notes, and there was like AIOps, platforms, different—what's the buzzword for you guys?

(Thapane (Deep Patel) at 00:17:11) That's a great question. I think AIOps is certainly this category. But if you think about, let's say, a Datadog or even AWS to a certain degree, they do a little bit of everything, and the word is more used at the application layer or the DevOps layer, where you're trying to deploy an application on multiple different servers and you apply monitoring and reporting and things like that. That's where the word has been used, but there's no reason—if I walked up to another CTO and I use the word AIOps, I think that's where their brain would go.

(Joel Beasley at 00:17:47) Where I would go. Yeah.

(Thapane (Deep Patel) at 00:17:48) Yeah, exactly. So some people have used it in our space. I think it's a little bit incorrect, but I don't think there's a buzzword. But I would say human-in-the-middle AI. Right?

(Thapane (Deep Patel) at 00:18:01) Because I don't think we're ever going to be able to completely get rid of humans. There's just intelligence required to solve these complicated problems where you don't have enough context that you can train a model on to be able to automate this process. So human-in-the-middle AI is, I would say, probably a buzzword for us. One thing is very clear that we're not going to solve this problem through conventional methods, and so we will have to start—we have trained models, and I mean, that's the only way this is going to work out. AI and ML are buzzwords, and everyone uses it. And, hey, we have this little magic dust sprinkled on top of our product. But there are very few companies out there where the technology itself is the product. Like, without it, the product simply cannot exist, and Trextel is one of those use cases. Like, if you think about Stitch Fix or similar companies, their entire model is, unless there is AI, some algorithms doing recommendations for you, there's no business. And Trextel is kind of in a similar space as we move more and more towards the software-as-a-service model rather than the managed service provider space.

(Joel Beasley at 00:19:11) Yeah. What's the—that's a good start. Like, what's the future? What's ten years out for Trextel?

(Thapane (Deep Patel) at 00:19:18) Yeah. So I think it's probably going to continue to move in this direction where, you know, I think we live in the world where NOCs should be ultra efficient, very small, and should always know what the resolutions are. And the only time when humans get involved is when something is totally out of scope of AI to be resolved. One example I give is currently we have models that can predict when the device is going to go down within 85 to 93% accuracy based on the previous performances of similar devices, that location, the type of circuits they're using, who the OEM is, what version of software they're using. And there are a couple hundred features that we mine.

(Thapane (Deep Patel) at 00:20:06) Based on that model, it says, okay, in next two to three hours, I'm 85% certain this device is going to have some sort of issue. So we can apply proactive remediation of such devices. And I think in ten years, the future looks more and more like this. You know, you can do simple things like rebooting, but in a complex multi-WAN—and when I say multi-WAN, it's like you have a Comcast connection, you have AT&T connection, you have wireless connection, you have satellite connection.

(Thapane (Deep Patel) at 00:20:34) How do you manage this at a retail location to have high bandwidth and high uptime availability of connection? Some folks have tried this at the hardware layer. So there are SD-WAN solutions that you can buy out there today. I'm going a little bit off on a tangent here, so pull me back.

(Thapane (Deep Patel) at 00:20:55) But this thing that has been tried at the hardware level, it only works like—so Cisco has a multi-WAN solution. Only works with Cisco. What about the others? In a large retail environment, you'll have multiple different equipments and systems that have to work together. So that's where a company like Trextel comes in and provides the sort of single pane of glass that can work with different manufacturers' systems.

(Thapane (Deep Patel) at 00:21:20) No matter who you have, it just works. And that's kind of the future. So if you are, you know, Walmart and you have 5,000 locations, imagine if you have one ticket each store. It's 5,000 tickets per day. I mean, imagine the human labor required to manage that.

(Thapane (Deep Patel) at 00:21:40) So we want to be able to reduce that, you know, hopefully to zero someday. But I think in ten years, make your NOCs much, much more efficient using this kind of proactive remediation.

(Joel Beasley at 00:21:52) All right. So I've got two things for you. The first one is, it's like when a new provider like Starlink comes in because you're the single pane, right, for all of these multi-WAN, like all these different providers, how do you interface with Starlink? Do they let you know a product's coming? Does it really matter?

(Joel Beasley at 00:22:09) Like, how do you start making sure your systems are compatible with that?

(Thapane (Deep Patel) at 00:22:15) Yeah. That's a great question. So in networking, there are a couple of protocols that are standard protocols. So just like for accessing websites, you use HTTP and HTTPS. There are similar protocols called SNMP and ICMP, and those protocols are common amongst network hardware manufacturers.

(Thapane (Deep Patel) at 00:22:36) These protocols were defined by IEEE many, many years ago. I think it was in the seventies or something like that. So really old protocols. They have been amended and so forth to work with modern equipment. But there's a standard way of telemetry collection, data collection from these devices and to be able to issue it commands. So it doesn't really matter when a Starlink comes around for us to have to be in the know.

(Thapane (Deep Patel) at 00:23:03) And I'm saying this in isolation. Of course, there are many, many benefits to partner with them and have sort of these API connections where systems can talk to each other. But from purely a technical standpoint, if there existed a network device and we needed to monitor it, theoretically, it's possible over ICMP and SNMP because those are just the standard industry protocols that everyone builds on.

(Joel Beasley at 00:23:30) And then are you seeing—like, have we actually seen deployments of Starlink yet? I saw that they had a few out.

(Thapane (Deep Patel) at 00:23:36) Yeah. So I personally know of a couple of deployments in California where they have been experimenting with it. So a couple of people that I know who have gotten it—just because they are around the Elon Musk globe—you can watch Netflix fine. It works. There are latency issues sometimes.

(Thapane (Deep Patel) at 00:23:58) So, but if you really experience it, you can't really tell that this is coming from a satellite. It's the general feedback that I have gotten. I don't know much more details in terms of the technology that they use and whatnot. But from a consumer standpoint, the people who have used it have really positive things to say so far.

(Joel Beasley at 00:24:19) We actually had our production meeting today. And one of the initiatives I gave one of our co-producers was like, I want to contact Starlink, which is owned by—if you go to their website, they're owned by SpaceX. So like, contact their PR people, tell them that we want to record a podcast episode over Starlink, like maybe with one of their engineers or something, and then have them send us a Starlink and we'll just use it for the recording and send it back. And then that way it'll be like a promotional thing and it'll be cool. And I don't know. And then I was thinking to myself, like, why am I so excited about that?

(Joel Beasley at 00:24:51) Like, I was like, I was such a nerd, man. I just want to record a podcast over Starlink. Okay.

(Thapane (Deep Patel) at 00:24:58) I think it's pretty cool. Right? Like, I mean, think about the bandwidth, the amount of—just I don't know how they manage the latency. That's the technical part that I haven't been able to wrap my head around. I think it's like sub-200 milliseconds, which if you think about like same distance like transoceanic, it's a little bit less.

(Thapane (Deep Patel) at 00:25:16) But to be able to maintain that over this—like you're going, I think, it's like a hundred miles. I don't know exactly the distance, but like a hundred miles or something in space and come back, beam of light, that is magical. Right? And we all use satellites today. Right? Like, you turn on Google Maps and, voila, you're using GPS.

(Thapane (Deep Patel) at 00:25:35) Right? So it's not that consumer use of satellite is new, but something about like streaming Netflix or recording a podcast over Internet streaming beaming from the satellites, it is exciting. I'm in the same boat. I would say that would be pretty cool.

(Joel Beasley at 00:25:52) Yeah. Because it's new. But my second question is about you mentioned like 5,000 tickets. Right? And the humans responding.

(Joel Beasley at 00:26:00) And so I'm curious, like, within your organization, have you made a special team that looks and groups the tickets by commonality to figure out which ones to AI-assist first or help automate? Like, how did that problem dictate your team structures?

(Thapane (Deep Patel) at 00:26:19) Yeah. So a lot of this today is dictated by how contracts are organized. So there's generally tier one, tier two, tier three. So think about a standard support team. And I talked about the broadband aggregation piece.

(Thapane (Deep Patel) at 00:26:37) So if you think about all the players involved in resolving a ticket, so there's a customer, there's a tier one, tier two, tier three MSP provider like us, there's broadband aggregator, there's the actual circuit provider, and there's the guy who does the billing. So there's seven people I said here. And depending on how the product or the service was purchased, everyone will play a different role. So, for example, let's say Trextel is tier one for a customer, which means our phone number is one of those 1-800 phone numbers that gets called, or our system is the first one that generates a ticket and our people start responding to it. Again, depending on how it was purchased, somebody else may have received that first phone call and then it moves towards us.

(Thapane (Deep Patel) at 00:27:28) But we are starting to build systems that—we have systems that connect these seven dots together and provide a holistic view of, okay, this is where the ticket originated. This is how it has traveled, and this is the final resolution, Mr. Customer. Your issue has been fixed by XYZ person. So in this crazy world of the movement of tickets and who's handling what specific part of the problem, it's really hard to—and I'm not sure if your question was with the lens of how do you organize this dataset from a training standpoint.

(Thapane (Deep Patel) at 00:28:05) But even from like human organizational standpoint, generally, when you deploy these services, your call center software, your support systems, sort of single pane of glass software gets configured for what you have purchased and how we sit in that stack of, you know, seven or eight moving pieces. So that's purely from a servicing standpoint. From a model training and sort of articulation of dataset, we treat every device as a node that's common. Like, it doesn't matter whether we are service one or tier one or tier two or, you know, how it was sold. But we say you're a device capable of doing Wi-Fi, and you have three interfaces, and you can do, let's say, Wi-Fi as well.

(Thapane (Deep Patel) at 00:29:01) So that's one category, and then all the traffic that comes from you as the device is being looked at in that kind of same category.

(Joel Beasley at 00:29:09) Yeah. Got it.

(Thapane (Deep Patel) at 00:29:10) So—and did I answer your question?

(Joel Beasley at 00:29:11) Yeah. We're getting there. All right. So you can look at it from the device specific level across all your companies regardless how the sales arm of the company happened. How do you prioritize or identify opportunities to generate automation? There's a million things you could do. How do you determine which ones are the ones you're going to work on now?

(Thapane (Deep Patel) at 00:29:35) Yeah. That's a great question. So depending—I'll give you one example. A lot of our customers tend to be retail, and there are generally similar types of devices that get deployed within an industry. So that's one way we have grouped them, or some similar type of devices.

(Thapane (Deep Patel) at 00:29:56) So, for example, let's say we're going to solve issues of APs because in the last one or two months, you know, that's been majority of our traffic for incoming calls or tickets that the systems are generating. So it is a little bit reactionary today where we look at the recent traffic of where the pressure is building from and then how do we solve that problem. But it generally self-categorizes itself where the loudest sort of gets served first and whether that's a device type or a specific industry type. For example, one of our customers, one of the largest banks in the world, because of COVID, majority of their users were, you know, sitting in their call center, in their properly set up systems behind firewalls and the secure environment where they're servicing their customer. And overnight, that changed.

(Thapane (Deep Patel) at 00:30:57) So, of course, that had to be prioritized over, you know, some of the other projects that we were running. So it is reactionary and the need of the day. But if I look at over, let's say, one year—I've only been here four months—but if I look at over one year how the systems have been prioritized, it's generally based on whatever section is the noisiest part of the network.

(Joel Beasley at 00:31:28) Okay. Yeah. That makes sense. And then it's also—it's a good thing that it's really clear what to work on. Right?

(Joel Beasley at 00:31:37) I always say whenever problems happen and the solutions are really, really clear, those are some of the greatest moments because there's really no other options. It's just like, yep. That may be a difficult thing, but at least it's clear. Right?

(Thapane (Deep Patel) at 00:31:51) Yeah. And I think, you know, this is a great lesson I'm taking from this conversation that figuring out metrics where you can be more proactive instead of reactive. In the world of MSPs, a lot of the times, it is the fire of the day, and so it's hard to like break that pattern and say, all right, let's zoom out and look at, you know, collectively what things need to be prioritized. And it's whatever is the buzzword of the day, you go and sort of chase that down.

(Thapane (Deep Patel) at 00:32:26)
Having said that, at the end of the day, these are—and I'm repeating myself here—from a technical networking node-by-node standpoint, there's some commonality between the problem of six months ago and the problem of now. So you are constantly moving and getting better and better and better, whether that's your model training or your operational procedures or things like that. And even though it seems chaotic, you continuously get better. Does that make sense?

(Joel Beasley at 00:32:56)
Oh, yeah. Yeah. In my mind, I try to paint a picture for myself, and I was wondering, do they have a team of people sitting around? And the reason is because I'm personally interested because of the fact that I've built two different companies historically and sold them, and they were both geared around automating or semi-automating human processes. So one was a financial software where this very expensive financial analyst was needed. And so to rapidly scale a company to multiple locations, you would have to hire twenty, thirty of these very expensive people, and then that made the business model not work as well.

(Joel Beasley at 00:33:37)
So I came in, worked with them for three years as a business partner, and built them things that made it efficient. So now you could have one of those people that could do the work of ten. So you really only needed two to run twenty locations. Right? And that made it all work. And so I did that, and then I did another one in real estate where we helped reduce the size of the number of people you would need on staff and things like that. So I was always curious—when you started talking about, "Oh, I joined this thirteen-year-old company, they've got these people and they're doing ticketing and automation," and I'm like, how are they—there must be so much opportunity. Like I said, kid in a candy shop. How are you prioritizing this? And you're like, "Well, they just really need us really bad. So we've got a lot of opportunity and we're just executing," and that's perfect. Right?

(Thapane (Deep Patel) at 00:34:25)
Yeah. I see where you were going with this now. I think in addition to the noise of the day, because there's a lot of low-hanging fruit—and I guess that's a better way of describing this—there are about twenty-nine to thirty resolutions that you can apply to a ticket based on which equipment it's coming from. That's a lot. Right? I mean, if you look at some of these operating procedures, the first thing you do is reboot. If that doesn't fix the problem, go on to this thing, then go on to this thing, then go on to this thing. And these manual processes are so well-defined that that also provides a little bit of a structure to go and say, "All right, we're going to figure out a system that can predict and reboot a device automatically. And if that doesn't fix the issue, then we're going to go and try the next thing."

(Thapane (Deep Patel) at 00:35:19)
So the class of problems that we may choose is, I think, random at the moment or noise of the day, but the application of method, I think—which is what you were asking about—comes from these predictive resolutions that a human applies. So the algorithm is, as a human, you go through this series of steps, and now we are trying to codify that whether it runs in carbon in humans or in silicon in computers. That's the process.

(Joel Beasley at 00:35:55)
I love it. I love it. So you're at this company. You're excited to be there. What's the thing or the project that you're really pumped up about today?

(Thapane (Deep Patel) at 00:36:05)
Yeah. So the project that we're working on, one of the examples I gave was this 85 to 93% predictability with which we can say this device is about to go down. If you think about that, that's kind of magical. Right? How many things can be predictive with that level of accuracy within that short period of time? And this is because we have terabytes of highly structured data that we can run models on to be able to achieve such a high level of predictability. Now this is in a lab environment, so I'm really excited to kind of dogfood our own system and launch it in early Q1 in our own NOC center to see how this is performing in the real world. And of course, the numbers are going to be a little different, but I was just so surprised when I heard the numbers and I saw the first set of results come in from the model training process. I never really expected it to be that good. I was like, "All right, if we get to 70%, that would be pretty fantastic, and we can just tell humans, 'Hey, there's a 70% chance this device is going to get into trouble.'" That would be pretty amazing. It'll make people more efficient and have a lot more context around the issue that they're trying to work against. Now that I'm starting to see these numbers, it's very, very encouraging.

(Thapane (Deep Patel) at 00:37:30)
Along with that, the other project that we are working on—and I will put them at a similar level of excitement—is anomaly detection. So because our business, or service providers in general around networking, so much of that is inbound traffic. And if you can figure out how to categorize the traffic on the fly, apply automated resolution, that just makes you so much more efficient. A lot of the times, one-off events happen in a network or at your home. Like, think about, hey, your power just flickered for a second, so it took your router down, and now it's back up. Like, that's a one-time thing. It's never going to happen again. But to a human or to a system, they don't really know the difference that the power flickered, which is why we're seeing this ticket.

(Thapane (Deep Patel) at 00:38:20)
So can we create a model that detects these anomalies and says, "Hey, this ticket, we logged it. Great. It's never going to happen again. Don't worry about it. Don't even look at it." Right? How can you get to a point where we are able to detect these anomalies? With enough amount of data, you can. And I'm starting to see some encouraging numbers. So it's a coin toss today. The model that we have in a lab environment is about fifty-fifty, which is fairly useless, but we're making great progress. We started with 2% accuracy. So from 2% to 50% is rather encouraging, and I'm really looking forward to getting those numbers to be in the nineties as well.

(Joel Beasley at 00:39:04)
Dude, that's so cool. Like, the predictive ability. My mind's just going off on that. Like, you could do so many unique things. Like, if you knew how many devices, you could do preordering projections and stuff, like, at what times of the year.

(Thapane (Deep Patel) at 00:39:19)
Exactly.

(Joel Beasley at 00:39:19)
I just saw some interesting data the other day. It was—I don't know why—but they were saying there's like a spike of 911 calls at 3 a.m. in the winter. And I was like, "Well, that's interesting. Why is that?" And it's because when people get out of bed to go to the bathroom at three in the morning, which is a common time for people apparently, the floor is very cold and your feet hit the floor and that constricts your blood vessels, and people have heart attacks. Like, older people have heart attacks. So there's actually a direct correlation in 911 calls in the winter at 3 a.m. to heart attacks and all of that. And I was like, "That is just—that is crazy." So I texted my brother-in-law, who has some heart valve thing. He's younger, but he has a heart valve. I was like, "Dude, get heated floors. What are you doing?"

(Thapane (Deep Patel) at 00:40:07)
I was like, "Get heated floors, man. I don't want—"

(Joel Beasley at 00:40:07)
Like—

(Thapane (Deep Patel) at 00:40:12)
Or, yeah, sleep in your bed with your slippers on or something. Wear some socks. Man. Yeah. But that's the kind of random thing that normal human beings don't think about. You come across these things as anecdotal evidence, but there's patterns that you can detect to say, "You know, yep, we're going to need more people around wintertime because 3 a.m. heart attack calls are coming in." Right? Like, there's so much—if you knew that statistic, that just makes sure—one of the 911 operators, imagine—there's lots you can do with just that one little statistic. Right? And that's what we're trying to do for networks.

(Joel Beasley at 00:40:54)
I love it. I love it. And when you came in—I mean, you said four months now—you came in, you've got the startup experience, you join an existing company. What was the culture like? Was there resistance to your style? Like, how did that go?

(Thapane (Deep Patel) at 00:41:09)
Yeah. And I would say we're still in the middle of it, figuring this out. And to be perfectly honest, I told Jason as we were discussing me coming on board, "Hey, you know, what excites you about this?" And this organizational change was equally exciting to me. I would rather say more exciting to me than just the software building part. Because, as I mentioned, you can go do software building anywhere. There are lots of different networking—like, if I want to work with networking, I can go work at any networking company. Like, why Trextel? It is this transformation of a business that has operated in a certain way, and now they're looking at this problem with a completely different lens. And trying to do that in a very rapid way. Right? So humans are not built for this kind of change, but the support that I've received from the senior leadership and just everybody in the team has been fantastic. Everyone's open-minded. Everyone understands that this is new for us.

(Thapane (Deep Patel) at 00:42:09)
Generally, what happens—if you think about a startup or any new business, right—you hopefully have figured out something that people want, and now you have to figure out distribution and then scale the company. That's generally how their journey goes. With a successful company like this, it's a little bit reversed. They've figured out distribution through the AT&Ts of the world and been able to grow revenue around $30 million, which is fantastic. But now how do you make the product in a way that can just put this thing on fire?

(Thapane (Deep Patel) at 00:42:48)
And so that was rather unique. Coming back to the culture, it is evolving. Sometimes I walk into a room and I feel like—I'm assuming here—people see me as this alien speaking a foreign language that they don't understand. But there's genuine curiosity. Like, people are, "All right, so when you say product-market fit, like, what do you really mean? Like, walk us through this." And so that opportunity to serve people and pull the whole team together, because everyone wants to get there. Right? The objective is very clear. Jason's done a great job of priming the team: "Hey, this is the objective. This is where we are going." Right? Putting the right people in the right seats. And so my job has become a little bit easier in terms of explaining people how we are going to get there, and we're not fighting whether we should go in that direction or not. If that was the fight, honestly, I would have struggled a little bit. Because if people just don't want to get somewhere or they don't understand what that thing is, it just feels very hard to fight that sort of headwind. But I think I got a little bit lucky.

(Thapane (Deep Patel) at 00:44:05)
I'll give you another anecdotal example. So we're talking about how do we test a new beta product that we are about to launch in early Q1. And, you know, I walk into some rooms and people start to ask me questions. "All right, so what's the plan? Like, how are we going to—like, I need a presentation." Right? Like, that's what they're trying to tell me. It's like, I can't create this presentation for you, unfortunately, because it's not this linear process where you build something, go sell it, and you have money. Right? It's more like defining how you draw a painting, because, you know, you have this perspective on the table. Maybe the customers don't buy it. Like, you're completely 100% off. I don't think we are that much off. We're going to be off. Right? So it's these series of pivots that you make towards building something. There are operational gaps. Right? So when I provide this answer, it's like, "Oh, but we need to train our teams on this new thing. Like, when do we train our teams? Like, how do you—you have to tell us. Like, this is going live on such and such date so I can train my product team or my support team," which is a totally valid question. Right? And like, how do you mesh this art and creative process with running a company? And so I think we're in the process of figuring this out as a company to make these worlds that generally don't mix together and make them mix.

(Joel Beasley at 00:45:47)
Yeah. And I know what you're saying too. It's a very hard thing to articulate. And myself having been through it, you know, launching products and figuring out—and then when people will ask me, "Oh, like, how do you do it?" And I gotta go high level. I gotta be like, "You gotta—you start and you don't give up until it's done." Like, that's the—if you want a formula to put on your wall, if you want one sentence, it's don't give up, don't quit. Like, you just adapt and figure it out. And I mean, there's some principles. And that's—you know what, as I'm getting older and more experienced, I'm learning the value and importance of principles because they are the concrete things that you can lean on, that you can build upon without knowing the specifics of the situation.

(Thapane (Deep Patel) at 00:46:32)
I see. So are you talking about like a Ray Dalio type principle? Like, what's a good example?

(Joel Beasley at 00:46:40)
Yeah. So from everything from Ray Dalio, like, about the customer, like focusing on the customer. It's as cliché as it is. They're the ones that are going to hand you the money. Right? They're the ones you're bringing value to. Understanding what value is and how—like, what is a Bitcoin or what is a dollar or what is a computer. They're these concepts of value, but I'm getting off for a second. But the principles concept, I found, tend to be the most important, you know, like with your team and trust and like who we are as a people because it's very much akin to the garden example, where I can't make the plant grow, but I can create an environment where good things can grow. And I don't know exactly what the plant will look like, but I know I can grow a certain style of plant. And so I need to put those seeds there. And then patience. Like, so all of these things—trust, transparency, patience—these things end up being the most important things because you can drive yourself mad as an entrepreneur trying to micromanage everyone and extending yourself, extending yourself. And then you finally get burnt out long enough that you realize you just have to find great people that are going to yield something positive and then just get everyone sort of on the same page and focused.

(Thapane (Deep Patel) at 00:47:56)
Wow. That's beautifully put. Like, you can create an environment where plants would grow. That's exactly the right kind of analogy. It's sometimes in the day-to-day hard to translate that. Like, if my COO is asking me, "Hey, when is this going live so I can train my people on it?" if I give that answer, I'm going to get punched in the face. Right?

(Joel Beasley at 00:48:18)
Give me—start talking about—

(Thapane (Deep Patel) at 00:48:22) Guardians. Exactly. Like, what the heck are you talking about, man? But that's the challenge, right?

(Thapane (Deep Patel) at 00:48:29) Like, I think people who know how to build and launch products, now you're trying to mesh that environment with something that's already working where there's structure. Right? So you're trying to infuse a little bit of controlled chaos in this structure. And most people are not comfortable with this, I think. Even founders sometimes are not comfortable with it. Right? But you have to do it. Right? And you have to tell people, hey.

(Thapane (Deep Patel) at 00:48:56) This is gonna be uncomfortable. There's gonna be chaos. You will feel like you don't know what the heck is going on, but it's part of the plan. So, like, trust the process, follow along with it, and that trust is built over time. And, you know, I'm sure parts of the organization today wonder, you know, hey.

(Thapane (Deep Patel) at 00:49:14) Why don't we have these clear answers and to be able to verify this in a way that they understand. Like, those are some of the interesting problems to me, not just the technology part, that just make me get out of bed and very excited every morning.

(Joel Beasley at 00:49:31) Oh, yeah. The communication thing. How do I deliver this so that it's accepted and not rejected? It's like, how do I hack around the human firewalls?

(Thapane (Deep Patel) at 00:49:40) Yeah. Right? Exactly.

(Joel Beasley at 00:49:42) Yeah. What are you learning right now? What are you as a leader learning right now?

(Thapane (Deep Patel) at 00:49:47) Yeah. So I had done a bit of, so my previous startup was in food tech, and we were delivering personalized dishes to people chilled, and so they just have to reheat and eat it every day. And I had, you know, I had to build a recommendation engine, but it was a very highly specific niche area of supervised learning that I got into. And I didn't know much about anything else. Like, I just read three or four books, followed some blogs, and people said, this is how we build recommendation and, you know, sizes like that because that was the need of the day, and didn't really understand how things under the hood work.

(Thapane (Deep Patel) at 00:50:29) And I'm starting to get a little bit under the hood for my own sanity because I go a little insane if I don't understand how something works. I think a lot of engineers are like that. So I'm starting to get a little under the hood. And then these communication things like hacking. Right?

(Thapane (Deep Patel) at 00:50:49) I didn't quite think about it that way until you said it, but it is that. Right? Like, so I'm doing micro pictures every day about specific things that I'm trying to communicate and see what's resonating. And then I hear people repeating the same or similar words to me about a topic that we had talked about earlier. I was like, yes.

(Thapane (Deep Patel) at 00:51:06) Victory. So that's been very exciting. And just networking, man, it's its own world. Like, there are so many, like, how does packet switching work? At a conceptual level, you may get it, but if you get into different manufacturers and different protocols that have been implemented just for Cisco.

(Thapane (Deep Patel) at 00:51:30) It's its own kind of world. It's like database systems. Right? If you try to understand how in-memory database works versus a document-based DB, like, it's just very different. Normally, users don't think about it.

(Thapane (Deep Patel) at 00:51:43) It's like you make an API call, something gets saved in the database. Well, it works. And same thing with networking. You don't really think about it until you have to solve the problem every day. So, yeah, these three things, lots to learn, a lot of little nuances to have to understand, specifically if you didn't come from this part of the world, like I didn't.

(Thapane (Deep Patel) at 00:52:06) I knew pretty much nothing about networking until I joined. Maybe I learned some things in school, and that was really it.

(Joel Beasley at 00:52:14) Yeah. And fresh eyes are,

(Thapane (Deep Patel) at 00:52:17) If you

(Joel Beasley at 00:52:17) have a really smart person, and then they get fresh eyes to a new industry, sometimes that can be illuminating. It can be very helpful.

(Thapane (Deep Patel) at 00:52:24) Yeah. It's like asking a lot of dumb questions. Like, hey. Why this? Like, why do we have to have static IPs for these devices?

(Thapane (Deep Patel) at 00:52:33) Well, so we can call them from the server. Well, that makes sense. But why can't the device call us? Right? It's just like, it's these asking a lot of dumb questions until you get to the bottom of it and annoying people.

(Thapane (Deep Patel) at 00:52:47) Like, okay. So explain to me how does this work? Why do we have this thing? That has been very rewarding. It's seldom you get a chance where you sort of get to climb down this mountain.

(Thapane (Deep Patel) at 00:52:57) Like, I built my career building mobile apps to start with, and then I have done some machine learning systems in the past. But, you know, those kind of mobile apps and sort of the back end applications is what I would say I built my career on. I recently, in the last four years or so, started building ML systems too. But it's just so difficult to climb down a mountain that you've already climbed on. Like, it's just so easy to say, okay.

(Thapane (Deep Patel) at 00:53:23) Mobile apps are my thing or websites are my, whatever that is, is your thing, and then you keep just going more and more in that direction. Personally, I've found it more interesting to climb down in one. Alright. I understand how that works. I may not be up to date on the latest and greatest library that does X, Y, and Z, but I'm gonna go learn this new thing from scratch that I know nothing about.

(Thapane (Deep Patel) at 00:53:44) It can be challenging at the same time, very rewarding.

(Joel Beasley at 00:53:48) Yeah. I found for my personal happiness, I need to be a beginner at something happening in my life. So whether, like, you know, this year I started really getting into piano and more music. Again, I'm already pretty good at the guitar stuff and I like being 101, at least in a small area of my life. Because if you're an expert in every single area of your life all the time, then you're not growing, you're not stretching, or you're not uncomfortable learning new things.

(Joel Beasley at 00:54:18) And that seems to be something like a muscle that you need to keep strong.

(Thapane (Deep Patel) at 00:54:22) Yeah. I mean, it just gets boring. Right? As simple as that. It gets boring after a while to keep doing the same exact thing in that way where, you know, someone walks up to you and says, I have this problem.

(Thapane (Deep Patel) at 00:54:33) Oh, here's how you do. I've done this 500 times. Like, this is the answer. Like, not having to do any work to be able to provide that answer all the time, it just, I can't live that way. No.

(Thapane (Deep Patel) at 00:54:44) It gets very repetitive.

(Joel Beasley at 00:54:46) This is great. So are you guys hiring right now? Are you hiring engineers and things like that?

(Thapane (Deep Patel) at 00:54:52) Yeah. We have multiple back end engineer and networking engineer positions. Go to trextel.com/careers if you're interested in learning about those opportunities. I'm on LinkedIn. I'm on Twitter at ITAP Android.

(Thapane (Deep Patel) at 00:55:06) Tweet me. Let's talk about your networking problems.

(Joel Beasley at 00:55:12) That's perfect. Yeah. And on your site, do you guys have a downloadable thing? Like, what's your bait that you use in sales?

(Thapane (Deep Patel) at 00:55:19) We don't have this material on the website, but if you go to the product summary part on the website, it talks about, it's not downloadable, but it talks about based on the industry that you are in. And of course, we can provide you a lot of different material if you reach out to me, Tapan Patel at trextel.com or just from the website, happy to talk about any kind of white paper or other case studies.

(Joel Beasley at 00:55:43) Yeah. What would be the thing that sets off in the back of their mind, I should reach out? Like,

(Thapane (Deep Patel) at 00:55:49) Yeah. Yeah. That's a great question. So no one wants to deal with NOCs.

(Thapane (Deep Patel) at 00:55:58) Even people who run NOCs don't want to have NOCs. So if you're one of those people who don't want to deal with a NOC or make your NOC more efficient, that would be the trigger why you would talk to us.

(Joel Beasley at 00:56:13) Perfect. See, that's easy. I like that. We could run ads with that all day. Right?

(Joel Beasley at 00:56:18) All these shady things. Like, frustrated with your NOC?

(Thapane (Deep Patel) at 00:56:22) Yeah. Yeah. Absolutely.

(Joel Beasley at 00:56:23) Slow NOC. We could do that. Yeah.

(Thapane (Deep Patel) at 00:56:26) Right? I love that. Slow NOC. Yes.

(Joel Beasley at 00:56:28) Slow NOC.

(Thapane (Deep Patel) at 00:56:30) Yeah. Yeah.

(Joel Beasley at 00:56:30) You know, play on words

(Thapane (Deep Patel) at 00:56:32) or something. We could do

(Joel Beasley at 00:56:32) some stuff. I'm auditioning for your marketing department right now.

(Thapane (Deep Patel) at 00:56:37) Hey. We are hiring in marketing too, you know, just so you know. You're good with words. I like that.

(Thapane (Deep Patel) at 00:56:43) Thank you guys. It's been phenomenal.

(Joel Beasley at 00:56:45) Alright. Talk soon, friend.

(Thapane (Deep Patel) at 00:56:47) Thank you. Bye.

(Joel Beasley at 00:56:50) Thank you so much for listening. And if you found this episode useful, please share it with a friend or colleague who you think would get value from it. And if you have topics that you'd like to hear discussed on the podcast, either add me on LinkedIn or send me an email, [email protected]. Every time I get an email or LinkedIn message, it absolutely makes my day and inspires me to keep going.