LinkedIn's Data-Driven Approach with Raghu Hiremagalur, CTO at LinkedIn

(Intro Narrator at 00:00:01) Today, we're talking to Raghu, CTO at LinkedIn, about being crisp when identifying problems and refocusing life beyond work. You're listening to the Modern CTO podcast.

(Joel Beasley at 00:00:17) I'm super glad that we get to hang out. Can you tell me a little bit about your background?

(Raghu at 00:00:22) Yeah, happy to. First of all, thank you for reaching out and setting up this time. Looking forward to this conversation. Joel, it looks like you have written more software for a longer duration professionally than I have. So I think this is going to be a very fun discussion. Me getting into my current line of work was actually not planned at all. My undergrad was in electrical engineering. So the stuff that I worked on was signal processing. And I worked quite a bit on computer vision related things in the late nineties. Worked on face detection, face recognition, a little bit of robotics, object tracking, et cetera. And I was in a PhD program trying to get a PhD in a related field, computer vision and related things. Halfway through my PhD, my lab mate and I, we started a company. It was a company that was trying to commercialize some tech that we had developed for motion tracking. Motion tracking stuff, I'm sure you've seen it with some Hollywood movies. They use very sophisticated motion tracking systems. We had some motion tracking tech that we developed. It was fairly simple. We were targeting physical therapists and chiropractors. So we started this company, we got into a business plan competition, and we won the competition. They gave us a bunch of money with no strings attached to go build this company. The company did not make much progress. We did not run it too well. But through that experience, one thing that became clear to me was that I did not know how to write software for a living and that I also knew that sticking to research, which is what I thought I wanted to do, was not my thing. So I decided to make a full pivot from whatever I was doing and even pivot from the field that I was working in. At that time, my wife, who was also in grad school with me, taught me object-oriented programming because I wanted to go interview with companies that were not in the computer vision field. Companies like PayPal or Microsoft or SAP, et cetera, et cetera. So that was my introduction to real computer science. So she got me interested in the area, taught me real programming, real object-oriented design, et cetera. Then I interviewed with a few companies. I was fortunate enough to get a job at PayPal with a team that was responsible for the core transaction processing system there. That was the first time I ever used a database. I practically did not know what a database was or how it worked or how to use it, but was fortunate enough to work with a set of people who brought me up to speed, taught me a whole bunch of things, mentors that I continue to be in touch with today. I was there for a little over four years. I decided to move on to go work in a gaming company. I joined Zynga. It was a very fast-paced environment. Had a blast at Zynga, experimented with all sorts of technologies. But through my experience at Zynga, it became clear to me that I wanted to work on things where I had deep passion for the product myself and had empathy for the users of the products that I was helping build. I'm not a gamer. So that became very evident to me. And as I thought about doing something else, through a series of conversations with one of my former colleagues from PayPal, I ended up at LinkedIn. I ended up at LinkedIn and I was hired here primarily to work on the feed platform. Feed back then at LinkedIn was still in its nascent stages. So I worked on that for a couple years, worked with a really, really awesome team to build our feed platform. Then again, through a series of conversations with my manager at that time, who happens to be my manager right even to this date, wanted to move a little further up the stack, wanted to work a little bit on the product side of things. And I had the opportunity to work on LinkedIn China. So figuring out how to open up access to the engineers we wanted to hire in China to LinkedIn systems was something that I had to figure out. And I share this frequently, one of the hardest things I've ever worked on to the state.

(Joel Beasley at 00:05:01) Why was this so hard?

(Raghu at 00:05:02) You know, it wasn't a technology problem. There are these concerns about opening up access to China. All sorts of concerns like, hey, how about data access? What level of access is acceptable? What is not, et cetera? And the setup we have from a networking standpoint with two layers of firewall, how do you provide the right access to code? Should we have access to production? All sorts of questions that we needed to answer, but we wanted to get to a point where engineers that we hired and put on the ground in China felt productive. We made reasonable progress with it and working through with legal, our security team, et cetera, that was interesting. So it wasn't a technology problem. It was ensuring that we got aligned behind what the goal was, what we wanted for our engineers there, and ensuring that we had the right controls in place but the right systems in place to make sure it happened. So anyway, we made reasonable progress on that. And around the time I was wrapping that up, the initial phase of that work, an opportunity to lead engineering for LinkedIn in Bangalore presented itself. I have no idea why I was offered that role. It was a big management job, and I had never really managed people at scale, at any scale really. It was a sizable team back then. I think it was around 350 people at the time I got to Bangalore. So the fact that I was even offered this opportunity, I took it. It was uncomfortable.

(Joel Beasley at 00:06:32) What did you learn from that? You just all of a sudden were thrown into managing 350 people. What was the big takeaway from that?

(Raghu at 00:06:39) You know, the biggest thing is, when people offer such opportunities, stretch opportunities, and if they believe you can do it and they're willing to stand behind you and support you, take it. Right? I took it and I got a ton of support, not just from my direct manager but from pretty much all of LinkedIn engineering leadership, which was awesome. So I moved back to Bangalore and it was meant to be a short-term stint, a year and a half or two. My wife continued to live in the US. She didn't move with me. So there was a little bit of back and forth of travel, et cetera, et cetera. We built a couple cool products out of Bangalore. We launched them. We launched them globally. And towards the end of my stint in Bangalore, I hired my successor there, moved back to the US, and then continued the conversation again with my manager about what I wanted to do next and what would be useful. And we started having a discussion about creation of a CTO role at LinkedIn. So we chatted about it for a while and he offered the role to me and I took it. It started off as an individual contributor role. Soon after, I had expanded in scope to not just being an individual contributor CTO, but to also lead a good chunk of our platforms and infrastructure. So that's the job I have today.

(Joel Beasley at 00:08:04) Do you ever get to work with Mohak?

(Raghu at 00:08:07) Yeah, he's my manager.

(Joel Beasley at 00:08:08) Oh, yeah?

(Raghu at 00:08:09) Yeah. I know you've spoken to him before.

(Joel Beasley at 00:08:11) Yeah, no, it was great. That's awesome, man. It's a small world, isn't it?

(Raghu at 00:08:15) That's right.

(Joel Beasley at 00:08:16) So what does your day-to-day look like now that it's evolved and you've been doing it for a little bit?

(Raghu at 00:08:20) It varies. Depends on the season, I guess. There are one-on-ones. There are technical decision discussions. There are strategy discussions. Again, depending on the time of the year, there are calibration discussions, promotion discussions. The one thing that I look forward to every week, we have around four hours set aside every Friday, my leads and I, where we spend time talking about just tech, about where we are in terms of state of the art, what needs to be built, why it matters, and how we can make progress towards building tech that can really create impact for the business. That's something I really look forward to.

(Joel Beasley at 00:09:03) Well, then tell me, what are your thoughts on the Instagrams and the TikToks? They've recently come up this trend in the past year with stories and these short clips. And when you as LinkedIn look at that, is that something that you think you're going to do? Can you not talk about it? Or what are your thoughts on that?

(Raghu at 00:09:21) You know, I'm probably not the right person to talk about it. But following those trends is, I don't think, our priority because what we are looking to do as a company is not the same as what TikTok and Instagram are trying to achieve. Creating economic opportunity for every member of the global workforce is what we are trying to do.

(Joel Beasley at 00:09:42) That sounds good.

(Raghu at 00:09:43) Right? So the products and features that we build are directed at making progress towards that. Now in achieving that, if something like stories, which is something we tried on the product, by the way...

(Joel Beasley at 00:09:56) Oh, you did?

(Raghu at 00:09:56) Yeah. We did. And we chose not to pursue it. We dialed it back down. So, again, I'm not the right person to talk about it, but following TikTok and Instagram is not what we're looking to do.

(Joel Beasley at 00:10:10) Right. And that makes complete sense. The stories and the shorts, I saw the trend. It happened. And then all of a sudden, you know, YouTube had them and everybody had them overnight. When I use LinkedIn, because actually LinkedIn is the platform that our company and me were the most active on because we do career advice, leadership advice, things like that. We found that at first, we didn't like the move where we had to sort of push things to a company page, you know, where it couldn't be off of Joel's personal profile, the API access change. But after we did that, I was like, okay, I see why we did that. That actually ends up working out really well because now I don't have 12 people logging into my personal one. They can be on company pages and manage it. Makes life much easier. And overall, I think you guys do a really great job, 100%. Not just saying that. I've noticed recently the algorithm got a little bit smarter where it started guessing other things I might want to be notified about. Because you want to see when your connections are commenting on things that you have interest in because then you can jump in and join the conversation. So I really like the direction that you guys are headed, and I enjoy using the product quite a bit.

(Raghu at 00:11:22) That's awesome. Great to hear.

(Joel Beasley at 00:11:23) So what's the next big thing for you and your team? What's going on at LinkedIn?

(Raghu at 00:11:28) So I'll tell you a little bit about a couple of things that are top of mind for me. So the product usage of the product and the value we create is accelerating. But what all of that means is pretty massive demand on infrastructure and platforms, accelerating demand on platforms and infrastructure. So ensuring that we have the right systems in place to support the growth and scale of the business is something that our teams have done remarkably well through the years. And we will definitely continue doing that, build new capabilities to ensure that infrastructure and platforms stay a couple steps ahead of where the business wants to go. But one thing that is sort of a new muscle that we are trying to build is to provide beautiful user experience for our engineers, for the infrastructure and platforms that we build for internal consumption. This is with ease of discovery of the systems that we have, ease of onboarding, ease of operations. This is not a muscle that comes naturally because we have exactly one customer for these systems that we build, that is LinkedIn. So it becomes easy to deprioritize building world-class user experience when you have a single customer. But it is really, really important because that to me is the single biggest unlock for productivity. When you have platforms and infrastructure that are easy to use, that are a delight to use, it improves the productivity of the entire organization multifold. So that is something that we're trying to push on. And this needs to be at literally all layers of the stack, be it with compute infrastructure, storage infrastructure, our caching systems, our database systems, all of these things. So that is definitely thematically an area that we are trying to focus on quite a bit. The second thing is figuring out ways to reduce overall cognitive overhead of usage of large-scale distributed systems. So, some context on what I mean by that. You know, n years ago, we all used to build applications on a dinky little database. It used to be vertically scaled to a certain extent, but most of your data used to fit in one OLTP system of some sort. You had a relational schema of sorts and used to read and write, and mostly the app used to work. But then the scale of all of these systems started to explode. The size of data started to grow, the complexity of data increased, and we started to specialize. We started to build OLAP systems. We started to build graph databases, and we needed fairly sophisticated stream processing systems. Overall, the surface area that gets exposed to application developers to achieve what they need to achieve just increased in scale. With that comes increased cognitive overhead, right? How do you ensure that a piece of data that you put into a document store somewhere flows seamlessly, gets transformed the right way and gets ingested into an OLAP system? How do you ensure these things are all working the right way? How do you configure the right streaming data pipelines for it? How should you think about mutations happening in one system flowing seamlessly into this other system and so on? So this level of cognitive overhead that we've created, I think there is an opportunity for us to reduce it to a certain extent. So pieces of infrastructure that don't need to be visible to our application developers, I think, should become invisible. Today, it is not. And this is something that we are trying to make progress on. I think it's hard, but I think it's a worthwhile thing to pursue. So this is something we are also focused on.

(Joel Beasley at 00:15:33) How do you do that? How do you find the right people to make these things easy?

(Raghu at 00:15:38) You know, the most important thing in making progress on that, Joel, is to have an outside-in perspective. What is your customer trying to do? I'll give you an example. Let's say if I want some sort of an insight using some sorts of data that LinkedIn has. And we use Pinot, which is an OLAP system that we built and open-sourced from LinkedIn. That's the system we use to serve up such insights. Ensuring that this piece of insight makes its way into the system called Pinot, it should be a couple double clicks into some sort of data catalog system where you say, hey, this is the type of insight I want, and these are the sources of truth of data. And everything that is required for that insight to come to life should be automated. Right? The data pipelines that are required should get auto-provisioned. They should start, data should start flowing through from the source of truth. The Pinot system should get provisioned with the right level of capacity and it should just flow. That is what I mean by it becoming seamless. It is the problem that your customer is trying to solve and ensuring that they get exposed to as few moving pieces as possible. So that is what we're trying to do.

(Raghu at 00:16:58) And the third area that we're trying to make a big push on is in ML or AI infrastructure. Scale and complexity of these AI models ever increasing and accelerating, but also fairly complicated in how you manage these systems. Getting to a point where we are able to expose AI capabilities as a platform where the entire life cycle of feature engineering, model training, model deployment, online inference, and health assurance, and the rinse repeat cycle of this whole thing is platformized, not just for small models, but large models where things just get hosted and managed is something we're trying to make a big, big push on. So that's something that I think about and spend a bunch of time on. The last thing is what we call data management.

(Raghu at 00:17:52) Probably goes without saying that this company runs on data. Right? And ensuring that we have the right guardrails in place to respect our members' privacy, comply with the right regulations in different parts of the world is really, really important to us. So building technology that creates the right guardrails to ensure we do right by our members and by regulations, while still creating the right tools and systems to provide powerful access to data for our teams is something that we're trying to push on with data management. It is hard.

(Raghu at 00:18:33) It is really hard to do this very, very well, but I think it's really important that we do this very well because this landscape is changing very rapidly. Ensuring that we have the right technology controls in place to do this well is important to us. So that's something we're spending a bunch of time on.

(Joel Beasley at 00:18:50) So all these initiatives, that seems to be where your focus is currently at LinkedIn. How do you get the right people to execute on this? What do you look for inside of the people that you're bringing onto your teams, involving in these projects? What sort of attributes do they have that make sure that, you know, you have confidence that you're going to put them there and the project's going to go well?

(Raghu at 00:19:12) Yeah. You know, over the last couple years, we have focused on both grooming fairly senior talent internally, but also recognized where we have gaps in senior technology leadership and hired people from the outside to plug those gaps. And these are people with deep expertise in these areas. So our job is to identify such talent, get aligned with what problems we want to solve in these areas, and let them do their thing. The thing that is important to me is that we get aligned on what the problem statement is, what success looks like, and just let them do their thing.

(Joel Beasley at 00:19:59) The problem statement. Earlier you said something—I think you said something about calibration discussions. I have never heard that before. What is that?

(Raghu at 00:20:06) Well, when I said calibration, I meant performance evaluation discussions.

(Joel Beasley at 00:20:11) It sounds cooler. It's an appropriate word because especially when you're working on a problem. Right? I found over building products that feature creep is a human trait. Right? You go off. You're like, okay, let's solve that. That would be cool. And then and and and. Right? And so being able to focus on that problem statement and doing only what's necessary to achieve the solution is super important. So when you said problem statement, I'm like, oh, I wonder if they talk about the problem statement in the calibration discussions.

(Raghu at 00:20:40) Yeah. I think it's really, really important to be crisp about what problem we're trying to solve and be very crisp about what success looks like. The thing that I insist on quite a bit when I get involved a little more deeply in a project is writing down what the design principles are.

(Joel Beasley at 00:21:02) What's the design principle? Give me an example of that.

(Raghu at 00:21:05) I'm going to repeat something one of my mentors at LinkedIn had told me a while back. Separation of the control plane from the data plane is really, really important. And all of that needs to be taken in context, obviously. So for a specific problem that you're working on, talking about what that specific principle means, I think is important. And the reason I think it's important to establish those upfront is because once you get into execution, it is easy for even fairly senior people to lose sight of what it is that you're trying to achieve.

(Raghu at 00:21:37) Because you're just so focused on execution that you start taking shortcuts that you really wouldn't take if you zoom out a little bit. Right? So codifying those things and ensuring that everyone working on a project or an initiative is aligned on those things and ensuring that we refer back to those from time to time is important. So as a stakeholder or as a sponsor of an initiative, for me, it is important that we get aligned on what success looks like and what the problem statement is.

(Joel Beasley at 00:22:05) And what's the interval on that? Is it just depending on the project and the person or is it every other week? Is it strict and restrained or is it just kind of flexible based off of the project?

(Raghu at 00:22:16) It depends on the project. I'll give you example of one more thing. For me, when you're writing a class, let's say a Java class, you should be able to describe the purpose of that class in a single sentence without an and in it. It does one thing and one thing well. If you start introducing ands in that statement, it means you haven't thought about it quite well. And I extend that to services as well. Tell me what that service does, and don't have ands in your sentence.

(Joel Beasley at 00:22:50) That is really good, and it's actually surprisingly pretty rare. I mean, I'm sure when you get into the higher end hiring, higher end developers with more experience, it might not be so rare. But in general, I don't think it's something that they're teaching in code camps yet.

(Raghu at 00:23:05) Yeah. And, you know, these things are hard to get right. But it's okay. I think these are things that we should aspire to get right as often as possible.

(Joel Beasley at 00:23:16) Oh, yeah. So change the conversation a little bit. What's your wife doing now? She's still writing object oriented code?

(Raghu at 00:23:22) Well, she's taking a career break. It's been a year since she started her break. She was a principal architect at PayPal. She was there for, I think, sixteen years. Worked on all sorts of things there. Yeah, she is on a career break right now.

(Joel Beasley at 00:23:37) Is she enjoying it? That's a very important part of life. Is she enjoying it?

(Raghu at 00:23:40) She is absolutely loving it. She gets to spend a lot of time with our two year old son.

(Joel Beasley at 00:23:46) Oh, nice.

(Raghu at 00:23:46) So that was one of the reasons she wanted to take a career break. But, yeah, she's enjoying it.

(Joel Beasley at 00:23:51) Oh, super smart move. My wife and I, we realized really quickly that it's better to take a cut in lifestyle and invest in the kids and be there and raise them versus, you know, because you realize when you're doing that whole childcare thing that you see them for an hour in the morning, really, and then an hour in the evening. And then eight hours a day, they're influenced by people that you don't know.

(Raghu at 00:24:13) 100%.

(Joel Beasley at 00:24:14) You can't get to know them on a level. They switch the grades so quickly. And so it's really just your responsibility as a parent to do that. And so we made that change, and it changed our entire life. We started homeschooling the kids. We sort of moved out of the city and moved out into a rural area. So we have a little farm now and everything. And we definitely were scared to make the move, but it was the right move. We're really happy with it now.

(Raghu at 00:24:39) That sounds amazing.

(Joel Beasley at 00:24:40) Yeah. But it's hard to do that.

(Raghu at 00:24:42) Yeah. And did you change your work style as well?

(Joel Beasley at 00:24:47) Oh, yeah. It was COVID that kind of made everything come crashing down where I had, so for the first two years, I was building, you know, the business, and then I was kind of ignoring my family a little bit. Then when COVID happened, the act of having to spend so much time and be so close together, it really forces you to either walk or commit. You know? And so for me, I think that was a good maturity point or at least, you know, a stage of maturity where I just chose to commit and lean into it, and I'm really happy that I made that decision.

(Raghu at 00:25:18) That's awesome.

(Joel Beasley at 00:25:19) Life is tough, man. Yeah. So what lessons are you learning right now in life?

(Raghu at 00:25:26) It's a hard question to answer, man. Yeah. One thing that I think about often, I don't have any siblings and my parents live in India. Okay. And my parents are older now. And one thing that I think about often is how far they are and how much time it'll take for me to get to them when there's a need.

(Joel Beasley at 00:25:49) Yes.

(Raghu at 00:25:51) And this is a choice that my wife and I have made to make this our home, this country our home and the Bay Area our home. So it's a trade off.

(Joel Beasley at 00:26:02) Are both parents—are her parents and your parents in India?

(Raghu at 00:26:05) Yeah. Both our parents live in Bangalore in India. You know, these are not things that we would even think about even a decade ago. But now it's something that is back of our minds almost on a daily basis. So ensuring that we give these relationships the right amount of time when it is possible is something that is really of lot of importance to us. And tying this back to what you said about the changes you made, these are changes that I have made after our son was born. My work hours used to be absolutely nuts. Now I make sure that I wrap up my day at 5 PM and go spend time with our son.

(Joel Beasley at 00:26:50) It goes fast. Yeah. People say that. Everybody says it. But then you experience it. And, you know, when you think of baby, for some reason, you think baby for a long time, but it's only like a baby for a year. Yeah. And then it starts growing. And, yeah, it's super important. Now for these life things that we're talking about, right, because I like that you actually went in this direction because this week, my wife and I were talking about that because we've been saying goodbye to our grandparents recently, and we're starting to, you know, in our mid thirties, and we're starting to sort of accept the circle of life. Right? And then, you know, try to set up our lives so that we do it well and that we take care of our responsibilities. But these human things, these life things, are these things that you'll talk about with your direct reports?

(Raghu at 00:27:37) Sometimes. Yeah. Honestly, these are not topics that are easy for me to talk about, but sometimes, yes.

(Joel Beasley at 00:27:43) They're hard topics. I find, though, it's very dependent on the person. Some people are more willing to talk about it. But from a performance perspective, one of the things that somebody had given me advice on, at one point in time a few years ago, this one specific employee, they're a super high performer, and then they dropped. I couldn't figure out why they dropped and they seemed distracted. And I was like, you know, everything's good with all our work relationships. They're happy with the money that they're making, but just something was off. And so, you know, I just started talking with them and it went into a personal area and it was appropriate and everything. And they were just struggling with something going on in their personal life.

(Raghu at 00:28:24) Yeah.

(Joel Beasley at 00:28:24) And I actually learned this trick too because I started doing that more often. I'll start doing that. And if it's something that's really deep, I'll actually connect them to a professional.

(Raghu at 00:28:35) That's really, really important. And thankfully, LinkedIn does a pretty good job ensuring that we have access to these resources when there is a need for people to talk to such professionals. 100%.

(Joel Beasley at 00:28:48) Well, we covered a lot of ground. I always like to ask as we sort of wrap up, what's the best leadership advice that you've ever gotten?

(Raghu at 00:28:57) So this was maybe eight or nine years ago now. I was working on something. I forget what it was. And there were a set of people who were unhappy with it, with what was going on, how it was being done, etcetera, etcetera. And I didn't hear about it directly, but I heard about it through other people. And I was all bothered. And I chatted with Mohak about it. He was like, dude, did you hear from me about it? Did you hear from me if I wanted you to change anything? The answer was no.

(Raghu at 00:29:29) He had not. And he was like, you know what? When you're trying to make changes, drive things, there will always be people who are unhappy about many, many things. Don't let that bother you. Right? There will always be a ton of noise. There'll be many, many naysayers, but you need to figure out how to distill the signal from the noise. So driving changes, building big things is not easy.

(Joel Beasley at 00:29:53) It is not. Yeah. That's good advice. I like that. What will LinkedIn look like in a hundred years?

(Raghu at 00:30:00) In a hundred years? Yeah. Wow. You know, creating economic opportunity for every member of the global workforce is a big, big task. Right? If you're able to do that for the global economy, I think that'll be amazing. If you're able to provide insights that help drive decisions on what kind of training people need in different geos to address shortages of labor in a specific industry, or to create upward mobility for people in different parts of the world for industrial trends. If we're able to drive those outcomes with the kinds of products that we build, I think that'll be amazing. I don't know if that's really hundred years out. That may be much sooner. But if that's the scale at which we're operating and those are the outcomes we are creating as a company, that'll be amazing.

(Joel Beasley at 00:30:58) So achieving your mission at scale.

(Raghu at 00:31:00) Yeah. Right? Yeah. That'll be cool. And for me as an engineer and as a platform and infra guy, building tech to power all of it will be amazing.

(Joel Beasley at 00:31:11) Right? Well, Raghu, man. We did it. We made a podcast. How do you feel?

(Raghu at 00:31:17) Awesome.

(Joel Beasley at 00:31:18) 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.