Scaling AI to Drive Revenue with Nishchay Shah, CTO of Cactus Communications

(Joel Beasley at 00:00:05) Hello, my friends. Today we're talking to Nishchay, CTO of Cactus Communications. And we discuss learnings from building several business units from the ground up, the challenges involved with scaling AI to drive revenue, and how to push hard and fail gracefully. All of this right here, right now on the Modern CTO Podcast.

(Joel Beasley at 00:00:33) This is the Modern CTO Podcast.

(Joel Beasley at 00:00:45) So tell me a little bit about your backstory, man. How did you first get into technology?

(Nishchay at 00:00:49) I was growing up, I was always a tinkerer. I was a very tech savvy guy. We grew up in a very humble household, but I always had access, a lot of books by our friends, family libraries. And I always read a lot and always got into figuring things out, how things work and getting into the innards of devices and appliances and whatnot. And then I got my first computer when I was in like sixth grade, I think mid-nineties, late nineties.

(Nishchay at 00:01:21) And it was like, whoa, mind blowing time, right? And then I started, there was a magazine that used to come to my house every, it was a monthly periodical. And in that it mentioned like HTML, the big thing. And I was like, whoa, what is HTML? And I started looking into it, and that's how I got introduced to the HTML at that time, which was basically the earlier version of JavaScript.

(Nishchay at 00:01:45) And then, yeah, and then I started tinkering with it. And I think eighth grade I built my first website and I remember very nicely it was something called tripod.com where you build your HTML, you do FTP upload, you put in your scripts in a certain directory and then try to, I think, vaguely remember, and then understand what DNS is at that point, a website name and whatnot. So that's how my whole tech thing started. But, you know, I grew up in a certain city where at that point hardware or electronics was like the big thing. So my undergrad, I chose that from undergrad.

(Nishchay at 00:02:26) I went into electronics and telecommunications. I had some subjects on programming, but not too many. But I remember in my third year, I worked on a very, very large personal project. We were asked to do something really small, but I was all ambitious. It failed drastically, but I was all ambitious, went into building a large project.

(Nishchay at 00:02:46) I had like microcontrollers and timing chips. And if I remember it correctly, the 8081, 8234, you know, all those chips, and built the hardware. And given that it was a very sophisticated at that point solution, I did a lot of software and that was written in assembly language. So at that point I realized that, hey, you know, hardware is really cool and great, and obviously it's necessary, but software is fast. You know, you can make changes and you can get that instant gratification, which you can't do that in hardware, right? I remember changing something means, you know, figuring out the whole circuit, soldering the electronic components, and then eventually it works. It's a long process.

(Joel Beasley at 00:03:33) Yeah. We had a guy on the podcast a while ago that was making AR headsets for firefighters.

(Nishchay at 00:03:40) Oh, very cool.

(Joel Beasley at 00:03:41) Yeah. It was really cool. It like outlines the physical objects in the room through the smoke so they can see what they're doing.

(Nishchay at 00:03:49) Oh, wow.

(Joel Beasley at 00:03:50) And he was talking about how it's really challenging to do hardware because he was like a, worked in software for a long time and moved into hardware for this venture. And he was saying that it's a big learning curve when suddenly moving a button by one inch costs you $100,000.

(Nishchay at 00:04:10) Absolutely. Yeah. No, absolutely. Absolutely.

(Nishchay at 00:04:14) No, I completely agree. Plus, it's software has a lot of wide uses, use cases. Right? Once you know some, once you're good at a certain language, you understand the basics of how things work, you have a wide variety of industry that you can contribute into. So anyhow, so that happened. After my engineering, I went to the U.S., did my master's. And there also I chose electrical engineering first and then moved to computer science because what happened was I was there and it was a two year course. And I realized that this is too long, you know? I mean, I have nothing else to do other than study. So let me do it faster. And then while that was happening, I started applying for the graduate assistantship or on campus jobs, but nothing worked out. And fortunately, at that point someone helped me saying, hey, you know what? Why don't you try programming?

(Nishchay at 00:05:12) And there are a lot of jobs available, and it worked out well. I learned a lot. I did this 30-day online crash course. And then I got a job when I was in the sixth month of my master's program. It was a full-time job, so I had to really balance it well. And I think that's how it started. You know, then I started as a developer, started with .NET code. Then quickly moving to web development because that was the hot thing, you know, in the late 2000s. And there I started at first I built a lot of tools as front-end, back-end. Then I started getting interested in how servers work, how back-end systems, database systems work. So then moved into systems engineering.

(Nishchay at 00:05:51) At that point, DevOps wasn't coined as a term. From there, moved on to various other jobs and then moved on to a boutique firm. It was a very small software firm, handful of people. And there I joined as a head of technology from like senior developer to directly head of technology. It was small, so obviously titles were fancy, right?

(Joel Beasley at 00:06:12) Yeah.

(Nishchay at 00:06:13) But what that helped me do is there I had to make my own decisions. You know, I had to make decisions on how something looks, where something goes, which tool to use, what libraries to use, guide people. I was also playing some of the product head role because I used to talk to the clients, listen to their requirements, design something. I learned how to do mockups, learned about the basics of user experience, you know, that it does matter. So, yeah, so I think from there I moved on to the world's largest, one of the world's largest event management software company.

(Nishchay at 00:06:45) And that was a great journey for me because after I came in, the business grew almost 500% in the four years I was there. So there I also worked as head of technology building products. I started having some product managers working with me. I built the DevOps function. We also grew a lot in 2008, 2009, 2010.

(Nishchay at 00:07:06) Cloud was the next big thing. And we were facing, fortunately, unfortunately, we were facing scaling problems, scaling issues. And we're like, okay, let's move to cloud. And I was the person responsible, the brain behind moving the full system to cloud. Requirement given to us was that since it's an event management software, a SaaS software, you cannot have any downtime.

(Nishchay at 00:07:29) It has to be zero downtime. And it's like, okay. Let's figure that out, how that works.

(Joel Beasley at 00:07:33) That's crazy. That's really cool that you were able to get to that, like, head of technology role at the smaller company and then still, and then just progress to continue in that at that level at the larger company. That's awesome.

(Nishchay at 00:07:49) Yeah. Yeah. Absolutely.

(Joel Beasley at 00:07:51) That's something I think about in, like, the trajectory of my career where, like, we're like a six-person company, and I have a lot of decision making that I do at our size. I'm always wondering, like, will anyone even care after this? Like, or do I have to go prove it and start from the bottom at a larger company if I decided to do that. You know?

(Nishchay at 00:08:20) No, I agree. I completely agree. And, you know, I always say this. So when I have younger gen, the next generation come in and chat with me, interns who come and chat with me or even some of my cousins, I always say that, you know, while the big companies are really good, there's Googles and Facebooks of the world and really good, definitely see if you can at some point in your career work in a small company in some significant role. Because what happens there is that you get hands on, you know, you get to make mistakes. You get to make really big mistakes. Plus, you get to know about how all the nitty gritty work, how things are dependent on each other and whatnot. Plus, for me, you know, I'm a citizen of India and I used to work in the U.S.

(Nishchay at 00:09:05) So there was this whole H-1B jazz and literally each and every friend of mine, for them it was essential that they get the green card. So they stay in one company forever, for a long period. Sorry, not forever. But for me, that never mattered. I don't care.

(Nishchay at 00:09:22) I need to make my career. I need to keep pushing, keep moving. And I don't really care about the green card as long as I'm getting really good work to do. And I'm happy, and I can see myself progressing. I think that's important.

(Joel Beasley at 00:09:35) Absolutely. So tell me how you met the team at Cactus and joined where you're at today.

(Nishchay at 00:09:42) Yeah. Absolutely. So towards the mid, end of 2014, just one fine day, we decided that me and my wife and then we have a daughter, that hey, you know, let's just move back to India. And it was just a spur of the moment. It's like, okay, fine.

(Nishchay at 00:10:00) Enough of the U.S. Let's move to India. And then I started talking to a few people, and that's how Cactus happened. And Cactus was a very fun, sorry, very, very, very interesting story because if you think about it, you know, when I joined Cactus, it was a team of 10 to 15 people in tech. And I was at a point in my career that I wanted to now grow significantly in all directions.

(Nishchay at 00:10:24) So it was a very big gamble, if you may, I took and it worked out pretty amazing because, you know, as I said, when I joined in, we had 10 to 15 people in tech. And today we have 300. And this is seven years. Yeah.

(Nishchay at 00:10:39) And it has been a fascinating and a very, very fulfilling experience and many more years to go. So Cactus, maybe I could talk a bit about what Cactus is. So Cactus, yeah, is a collection of many businesses, right? It's not really one business. So we have Editage, we have Cactus Life Sciences, we have Impact Science, PaperPal, so on and so forth. And we primarily work in the academia and the pharma space. And we are one of the very few large companies which are author centric, which means essentially what we do is we bridge communication gap between research and the world. We help the researcher, the author disseminate their research to the world, you know, starting from helping them with translation, the subject matter editing, providing them publication support and to the point, you know, helping the researchers promote their published research to the world.

(Nishchay at 00:11:34) You know, so we are a tech product company. We have many suites of products. So PaperPal is one of the suites, ResearcherLife and whatnot. And that's where, that's where for me, my focus started going into, as a large part of my, like, almost 100% of my focus was in building the team. Started with like a person and a half, like one and a half people, me and one of the devs. We started building it, a lot of failures along the way, but fortunately things worked out well. And today we have a machine learning group called Cactus Labs, which is generating a lot of IP for us, which is generating a lot of AI solutions, which then those solutions are being used by other businesses within Cactus to push their business up. And plus, many of these solutions that we've built are being packaged into products under the umbrella term PaperPal and ResearcherLife and whatnot. So, yeah. So that's how Cactus really happened.

(Joel Beasley at 00:12:44) That's awesome, man. So I know that Cactus started by, like, helping scientists get their papers and work published. Is that still a big part of the business today?

(Nishchay at 00:12:58) Yes. Yes. Absolutely. It is still a big part of the business, and it will continue to be a big part of business for the foreseeable future.

(Joel Beasley at 00:13:07) So do you have, like, a, so how does that work exactly? What is your role there? Do you provide, like, subject matter experts that, like, help with peer review or, um, like, what's the value add that Cactus does for these scientists that are trying to get published?

(Nishchay at 00:13:27) Sure. No. Absolutely. I think from a very high level perspective, if I have to say it, is that we provide the language support. Right? So as a researcher, as an author, you're good at your subject, you're good at your experiment, you're good at your research. But that research has to go out somewhere that people can have access to it. Right. So it's like you build a product, but that product is useless until you can market it in the right channels. Similarly, with the researcher, you know, for them, for the research to have any credibility, it has to be published into a journal first.

(Nishchay at 00:14:02) So that's where we come in. You know, we take the research. If it is not in the English language, we translate it from a native language to English language. Then we have editors who work on the paper to edit it. We have formatters who will format it for the journal requirements and then help and then send it back to the author who can then get it published into a journal.

(Nishchay at 00:14:24) The translators and the editors are the key components of the key value adds, right? Because if it's a paper on cancer, we will need to, it is essential or it's actually, it is mandatory that an oncologist is editing the paper because the subject matter matters a lot. So you cannot have an architectural major edit a social sciences paper or you cannot have an anthropology guy edit a neurology paper. It's just different subject areas, very different tone, language understanding.

(Joel Beasley at 00:15:04) That's really cool. So recently I actually got to interview an executive from this company called Sophos that they're a cybersecurity company, and they do a lot of really cool stuff stopping ransomware. The one thing that was really interesting about them is they actually have, like, this whole website separate from their main website dedicated to publishing scholarly papers on AI. And they, like, encourage their in-house experts to do research and publish it. He was saying that, like you were saying about Cactus Labs, encouraging your in-house people to do research on AI, obviously, like, creates a lot of valuable IP for the company.

(Joel Beasley at 00:15:50) And it also was, like he said it draws in, like, really smart people that are doing academic research that want to come in and join the company because of the opportunity to get published. And so I was curious. Do you guys help any of your internal subject matter experts get published themselves?

(Nishchay at 00:16:11) Yeah. So, unfortunately, we have not been very good at that because, you know, we are still growing. We are at an inflection point now because early on when we went to the business saying, well, we need to put money in AI, obviously, from any business angle, it will give us outcome. Right. And in order to push for the outcome, we have to, you know, focus on the outcome. So the last couple of years, which is the first couple of years of AI, we were focusing a lot on outcome, building a team, building the capabilities, and now we have started. So we already have an intern who's not an intern now. Now he's one of the key members of our team who actually wrote his thesis with us.

(Nishchay at 00:16:52) So he was in his program, wrote his thesis with us, got it published while he was working with us on our solution. And we are also currently looking for researchers who can actually do that. So I think the next couple of years will be that where we have acquired critical mass and now we will move into more research and actually giving back to the community.

(Joel Beasley at 00:17:14) That's really cool. So what are some of the cool things that you've seen come out of Cactus Labs and the research there?

(Nishchay at 00:17:22) Oh, yeah. So Cactus, I think when we started, right, the thing was that how do we leverage AI to make a difference in business? And we know that for 20 years, Cactus has been in the business of editing papers, editing literature. So we have collected massive amounts of data, our own IP, our editors making edits.

(Nishchay at 00:17:45) So we said, let's start using that in order to build an AI that can edit. So that is our number one contribution to the company and to the world where we have leveraged NLP, we have leveraged deep learning language models and whatnot to generate something called an automated editing system, which will take an unedited piece of text in English and emit a grammatically correct, structurally correct piece of text. That has been our number one win. And we've got many more. So we also have concept extraction where we take a piece of text and we extract key concepts, semantically correct concepts out of the text and create a user fingerprint or text fingerprint, which can then be used to map or match similar literature.

(Nishchay at 00:18:42) So that's what we build. We are a lot into big data today. So we are collecting all the world's intelligence. There's a better way to say it is that we're collecting all the world's published papers, whichever is available online, open access. And we are creating a big, big data lake and we are running a lot of intelligence on top of it.

(Nishchay at 00:19:02) And we have a lot of plans to run, again, a lot of AI on top of it to generate value for the business. So those are some of the things we've done. We also built some image recognition systems using YOLO and other libraries where we would detect birthmark status on human skin to check for consent. If you're using that image in a paper, you check for consent. So we built that.

(Nishchay at 00:19:25) And yes, we also tinker a little bit with AR. Again, that was more of a fun project, side project kind of a thing. It could become big, but we don't really have the bandwidth to productize it today.

(Joel Beasley at 00:19:37) That's awesome. So did you have an extensive background in AI when you started implementing it at Cactus, or how'd you get started with it?

(Nishchay at 00:19:49) Yeah, that's a very good question. So yeah, it was a very interesting journey. So I had, I have no background in AI, but fortunately I have done my master's in computer science. So I understand a lot of the basics of computer science, basics of math that goes into building AI, right? So when we started, the thinking was that instead of, you know, as a leader, you always have to make a decision. Should you buy or should you build, right?

(Nishchay at 00:20:18) And buy versus build is always a question. So when we did start, as I said, one and a half member team, the thinking was instead of building, let's just buy. Let's just work with a consulting agency who can provide us experts and help us build something. And that's what we did. That's what we started with. But that never worked out for various reasons because, again, it's an emerging industry.

(Nishchay at 00:20:42) So there are a lot of players in the market who are not really experts. They're more of statement experts. But so that didn't work out. So then we started working with universities, so we partnered with universities in the US and worked with their head of computational linguistics and worked with him and his team to build something. And when that was happening, for me, I had to do a lot of learning myself.

(Nishchay at 00:21:12) So, you know, reading papers, learning about multidimensional algebra, linear algebra, probability. You know, there are really good courses on Stanford and MIT on MIT OpenCourseWare on how probability and statistics work. So I did all of that online, you know, went through Coursera and whatnot just so that I can speak the language of the people that I hire, of the people that I work with. I personally haven't really written a lot of code. Earlier I did. Earlier we did because we were using something called PyTorch, which was a, which is an unsophisticated library.

(Nishchay at 00:21:56) So you had to really tinker around with architecture. But yeah, so that's how the journey came to be.

(Joel Beasley at 00:22:02) That's really cool. So what advice would you have for smaller companies that are looking to start building out an AI team from that experience?

(Nishchay at 00:22:13) Yeah, I think what matters is get a couple of right people. So we feel that early on, we hired a few people who were not right and that also, you know, slowed us down. But then fortunately, you know, we got one or two good people and those people really helped us. So don't fret on investing in the right people. It will be a large investment.

(Nishchay at 00:22:30) But let that happen. And also whoever's leading it. So any business leaders, I was a business leader leading it at that point. It is essential that it is not one of the many goals or one of the many targets that they have. You know, it is essential that for the business leader who's leading it, it is their singular focus, because once it is set up, you can definitely move to other things. But when you're setting it up, it is essential to have the right focus and continuous focus and absolute persistence. Very, very necessary because there will be many failures on the way. Making a simple AI, making something that works in test is very easy, but actually scaling it to production such that you're actually generating money out of it.

(Nishchay at 00:23:23) That is hard because it requires a lot of feedback loops and a lot of changes in workflow, you know, to accommodate AI. So AI is not a drop-in replacement. You have to accommodate it. You have to make sure that you are changing as a business in order to leverage it well.

(Joel Beasley at 00:23:40) That makes sense, especially just how difficult it is to productize because when you're just experimenting and trying to iterate and make it better, if there's failures, that's fine. You know, you just iterate on it. But once you're charging people money, then if there's failures, you've got to answer for it. And that seems stressful. So one thing I was curious to ask you because you were talking about working in different head of technology roles and now CTO.

(Joel Beasley at 00:24:12) How would you say that those two job titles are different day to day?

(Nishchay at 00:24:19) They are a lot different, right? So actually, the role of a CTO, the title of CTO does not really have a very strict global definition, right? And it should not. You know, it is essential that the title may remain the same, but the role has to keep changing. So as I said, when I first joined in, I was actually playing the tech head role and I was actually writing code. I was helping people select the right libraries to use, the right tools to use. I remember doing not one session, multiple sessions on how to use Git, how to version your code well, how to deploy it well, you know, stuff like that. So it's very deep down, hands on.

(Nishchay at 00:25:08) And then I moved into, you know, building various types of teams that are required for a successful tech company, right? So technology is one thing. Software development is one thing, but you need QA. You need automation. You have QA.

(Nishchay at 00:25:23) You need product management, you need user experience team. You need all these teams in order to build a successful tech company. So that was the role I was playing. Then, you know, while the title remained the same, the role I was playing was basically a business consultant to build a tech company. Then as we progressed, you know, I was playing the role of how to expand well.

(Nishchay at 00:25:46) So, you know, expanding comes in two flavors, right? One is you keep hiring people internally or you start hiring externally, which is you outsource. So I think one of the years I focused a lot on how to outsource. So now today we have seven different agencies that work with us where we outsource a lot of things, smaller parts. And setting that up was difficult because then you don't have to be a tech role.

(Nishchay at 00:26:12) I mean, tech, yes. You also have to play a business role again. You have to negotiate pricing. You have to negotiate contracts and so on and so forth. Then I think one of the other roles I have played is a sales role, a sales tech role or a sales tech integration role where I used to meet with the publishers. You know, I've traveled to almost all the top four publishers' offices, spoken with them multiple times online.

(Nishchay at 00:26:38) How can we integrate our software, our tools within their ecosystem? And why should we do that? Not so much about the act of integration, which is the tech guy's job, but also playing the tech partner role. So that is another role I played and I have played. And then finally, the AI role there, you know, building the function.

(Nishchay at 00:27:01) And there again, I went back to being hands on early on. So I was really hands on writing code, you know, going to forums and figuring out solutions, contacting the developers on GitHub, asking them for help on maybe stuff we can't deploy. Things aren't scaling. We aren't able to utilize the full GPU, you know, all those things. So, yeah.

(Nishchay at 00:27:21) So I think that's how the roles differ. And I always believe that, you know, at a leadership level, your evolution will make the company evolve and it's not vice versa. You know, you have to evolve and your evolution will basically push the company to evolve in the right direction. And that is very, very critical.

(Joel Beasley at 00:27:45) Sounds like in your time there creating all these different functions that you've talked about, most recently AI function, it's a cycle of being really hands on and then slowly stepping back.

(Nishchay at 00:27:59) Yes. Absolutely.

(Joel Beasley at 00:28:00) I imagine in those early days of starting up a new function when you're really hands on, failure obviously is a big part of it and a positive part of it because you're iterating and creating something new. But as a leader, when there is a big failure, how do you communicate that to your team gracefully?

(Nishchay at 00:28:25) Yeah. Yeah. So that's a term I use a lot at work. You know? If you do fail, please fail gracefully. It is not just okay to fail. It is necessary to fail because if you have all the wins all the time, you're not really building the right thing. You're not really pushing hard. You're not pushing hard enough.

(Nishchay at 00:28:46) You don't have the resistance. So it's okay to fail because then you're taking risks. So I've failed a lot in my life, right, starting from the first job, starting from studies, jobs, etcetera. And I have not always failed gracefully. And over the years, I've learned how to do that.

(Nishchay at 00:29:05) So as I said, you know, it's essential to fail gracefully. It is essential to be aware of what is happening around you. You know, it is essential to own your own failures. You know, it is very, very essential to own that. Okay, I failed, you know, take the responsibility of the mess and figure out how to clean it up.

(Nishchay at 00:29:24) Ask for help. You know, a lot of times it does happen that you don't ask for help. You know, you are standing in the middle of the road, you know, staring in the headlights and like, okay, now what do I do? So don't do that. You know, be aware and be mindful that you're not doing that.

(Nishchay at 00:29:42) You're asking, you're failing actively and talk about it. So fortunately, you know, at Cactus, we have the culture. We have a very good culture in terms of failures. It's okay to fail as long as you're learning from it, as long as there's reflection, you know? And so even Ray Dalio used to always say, and he does say, one of those principles, that pain plus reflection is progress.

(Nishchay at 00:30:04) So if you have the reflection after you have failed, that's where you make your progress. And lastly, I think it's very, very essential to make yourself aware. Understand what is in your control. You know, what is your circle of influence and where you can't influence, ask for help and figure it out.

(Joel Beasley at 00:30:26) That is excellent advice, man. Thank you for sharing that. So before we wrap up, is there anything that we didn't get to touch on today that we want to make sure we get out to the world? Anything you want to shout out for Cactus? What do you want people to remember?

(Nishchay at 00:30:42) No, I think from a leadership side, it is, what I would say is that, you know, as a CTO, as a head of technology, think of your role as a QA. You know, you are a QA. You bring in the vision. You help verify the vision. You help your second line, your third line, your team members commit to the vision. You know, and once they're committed, they'll have accountability. You give them accountability.

(Nishchay at 00:31:04) You give them the power of decision making and you eject yourself from it. You know, you don't become part of the design, don't become part of the solution all the time. And that's how I have seen the teams grow very well. When you're in a startup, that's the opposite, right? When you're a startup, you have to be part of the solution.

(Nishchay at 00:31:22) You have to be part of the design. You have to be hands on and make things happen. But at some point, you've got to scale up. And when you do scale up, you have to figure out how do you let it go and observe from 30,000 feet and you come in, you test it out and you move out. So I think that's very critical to grow, and that has always helped me grow the teams.

(Nishchay at 00:31:46) So even Cactus Labs, we started as a one member team. Now we have about 50 people. The reason we are able to do that is because, you know, even my second line, who are the VPs of the organizations, that's what I always tell them, just eject yourself. Don't make yourself part of the problem.

(Joel Beasley at 00:32:03) Yeah. Absolutely. So are you guys hiring? Should people reach out to you if they're looking to join your team?

(Nishchay at 00:32:11) Yes. Absolutely. We are hiring. We are hiring. If you are a tech person and if tech interests you, if AI interests you, front end, back end, DevOps, everything.

(Nishchay at 00:32:21) We're hiring for literally all positions, including product management, user experience, user research. Yeah. So we are looking to grow our team. We want to grow our team drastically in the next couple of years. So definitely please reach out to me.

(Joel Beasley at 00:32:40) 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 would 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.