In The Arena by TechArena - Inside a High-Stakes Infrastructure Overhaul

Episode Date: July 28, 2026

In this episode of In the Arena, host Allyson Klein sits down with Rakesh Awasthi, a principal software engineer, to explore what it actually takes to modernize legacy data infrastructure at global scale and how AI has fundamentally changed the nature of that work in a preview of his talk at the AI Infra Summit in September 2026.

Welcome to Tech Arena, featuring authentic discussions between tech's leading innovators and our host, Allison Kline. Now, let's step into the arena. Welcome in the arena. My name is Allison Klym. Today, I'm delighted to be with Rakesh Svati, Principal Software Engineer and MasterCard. Welcome to the program, Rakesh. How are you doing? I'm doing wonderful. Thanks, Alison. Thanks for having me here today. I'm really excited to be here, and I'm really looking forward to our conversation today. It's your first time on the program, and this is part of our coverage leading up to this year's AI Infra Summit,

taking place in September in the Bay Area. Can you just give our listeners to start a sense of your background and the kind of work that you do at MasterCard? Yeah, absolutely. So my name is Rakeh Avasti, and I'm a principal software engineer at MasterCard. And for last two decades, I've primarily been focusing on data engineering

and large-scale data platforms. And most recently, I've been leading efforts around building master data management solution and modernizing our legacy applications. I would say that these systems are critical because they are many customer-facing products across MasterCard. We're ensuring high-quality and consistent data at the scale is absolutely essential. You know, it's interesting when you consider MasterCard and the services that you offer real-time capability and reliability is so critical.

And I know that you have a ton of historic applications that drive services all over the globe. Can you tell us about how the landscape has changed and how you've addressed this through modernization of your applications? Yeah, of course. And I think one thing we need to recognize about these legacy applications is that these systems are incredibly robust, right? They've been built and optimized over decades to handle massive scale with very high reliability. What's striving this modernization effort now is that the environment around these systems is fundamentally shifting.

Now, there is a much stronger need for real-time payments, open banking, payment facilitators, binopulator, cryptocurrencies, cryptocurrencies, regulatory requirements last nearly seven, eight years. A lot of those have changed, right? So these, along with the need for more AI-driven use cases, right? So these things are actually, you know, shifting. And these are the driving factors why we have to modernize the systems, right? And these systems were, you know, not designed for these use cases initially. So there is a strong need.

And we also have to look at that modernization is not very straightforward. These are mission critical systems and there is very low tolerance for disruption. And these systems are deeply interconnected with a lot of implicit dependencies and domain knowledge has also evolved over time, which makes it very, I would say, inherently risky to change, right? So in practice, we have to strike a balance between modernizing incrementally while preserving the stability and reliability of these systems that they were originally designed for. Now, Kesh, I know that you're speaking at AI Infra, and without spoiling your entire talk,

I know that your teams have leaned into AI to deliver some of the modern capabilities that MasterCard is seeking. Can you walk us through how that journey has gone for you? Yeah, absolutely. At MasterCard, we operate at a very massive scale. And one of our core systems, like the process, transaction, data to extract and normalize merchant location information, I would say essentially what we are doing is we are taking very noisy, very unstructured data, cleansing it, and assigning a very unique identifier to every merchant location, which then powers a lot of downstream applications.

Over time, we have started seeing science of aging, and we have also seen the factors that are driving this modernization, right? So when we brought AI into the picture, we looked at AI in two different ways. One was that use AI to better understand our data? improved data quality, enabling better, smarter classification, detect anomalies, beyond what a rule-based system can catch. And second, which was more interesting, that using AI as a partner in modernization, helping engineers understand lineas, debug faster, and accelerate transformation journey,

instead of doing everything manually. Now, we also need to understand that the system that was originally written probably a couple of decades ago, many of the original data engineers are no longer with us, right? So things have really shifted. So we are not using AI only to just to rewrite or optimize, but we are actually using that as a partner. And I would say that that transformation is really the turning point for us. Now, I know that you had targets for AI in terms of what you were trying to accomplish, but when you started utilizing the technology, did it open up possibilities that you hadn't originally anticipated.

Yeah, I would say that before AI, you know, our legacy modernization effort might have looked very traditional. Read the documentation that is available to us, rely on our experience, the engineers intuition, pick up a technology stack which we think at that point in time

will best meet our need, and then spend probably a couple of years rewriting with a small team, right? And then in that world, it will be more of a exercise of technology selection and migration. I would say that what AI has shifted is like instead of it being a rewrite problem, it became more of a understanding problem, right? And instead of debating what would be the best tech stack for weeks and months,

we could actually very rapidly build a prototype. It helped us pick up the best language, best architecture for our need. So we were using AI to accelerate comprehension of what our system does, right, and answer a lot of those questions which were probably some of the undocumented H-cases or hidden dependencies and all that, uncovering that. So I would say that what could have been a longer journey, we were able to compress that and have it modernized not just the application, but actually a broader ecosystem around that because we were also able to build

the interfaces, workflows, or learning platform, and something that will actually evolve from here going forward, much better. So beyond the one, the projects that you're going to be talking about at the conference, how are you something about the larger picture? What have you learned about how AI is going to change the daily lives of data engineers? And that's very interesting. And I think that is coming quite often these days, right? Because if you really look at, especially after generative AI came,

how the narrative is really shifting in last couple of quarters or so. And I really don't see AI taking away all those jobs of data engineers and all that. I think what will happen is that role will get actually redefined. The center of gravity for that role will shift. Some of the work, like which is repetitive, for example, or exploratory, that will get accelerated. And the focus for the engineers will shift from there into more of understanding better system design or data quality, observality and judgment and all that.

One more thing will happen is that, and this is enough interesting, right? AI can generate code very faster, right? But then again, as a result of that, cost of going wrong is also going up. Being able to validate the AI generated output will become actually a required very core skill that data engineers will have to have. They need to know how to check outwards, how to understand the limit of AI model, right, how to use the tool responsibility. So what I would say is that the role is actually becoming much more strategic, not less. Now, fun fact about you, when you're not working at modernizing applications for MasterCard, you are a meditation instructor focused on mental health and wellness.

What does that training and focus in your life tell you about the human elements of integration of AI into the workforce and how teams should be grappling with the change and the impacts of change? I actually started my meditation journey at my workplace, which over two decades ago. And ever since it has been very important part of my life. I have been a trainer for nearly a decade now. And what I found about heartfulness meditation is that it is very practical. And you really learn how to stabilize your attention, how to create small pauses, and actually respond to the situation instead of reacting. Now, when you look at from the time that we are in, like from this AI era,

the need for meditation actually becomes much more relevant. Because of AI, you know, it accelerates the pace of change, right? You're going through so much of in so little time, so much of change, right? There is a technology shift. And it also needs to cognitive shift. So people are continuously dealing with the context switching, information overload. And I need to know how to move faster and handle pressure, right? I would think about like, you know, AI hygiene, for example, right?

In last 10 or 15 years, we have slowly learned about the digital hygiene. But now we need to learn. about the AI hygiene also because we need to develop our habits around that how to step back validate our output how to be intentional about what we want AI to do that and when to rely on it and when to use our or apply our judgment right not just letting the speed of the tool dictate the pace of our thinking and AI can be very addictive also so we need to handle that and learn how to use the tool in the right way and that is where if you look at the heartfulness Well, poor idea, clarity before action, I would say.

That is where meditation genuinely helps. No, I know your team is well along the journey of AI integration into your strategy. But if you're talking to folks who may be earlier in the journey, what would you say to them in terms of thinking about what's possible through AI integration? I would say first start with some real problem, something that you already have at hand, right? Don't use AI just as a concept, but pick up something around your own application. Start slow.

Look at the task which are repetitive in nature, right? Or something which is right now difficult to understand. And apply AI there. That is probably the fastest way to learn very quickly. And second thing I would say is that treat it as a very disciplined practice, not just as a productivity boost, but think about how to apply my human just. management, right, or where we should go into the critical systems and what our real customers want and how it will affect them or what will be the business outcome because of that.

There can be one more way we can look at is like look at how it can become a team learning process instead of individually, right? I think the organizations or teams which will stay ahead of the others would be like not the ones which have most tools, but the ones which will be able to build that shared understanding of these tools and their capabilities. I would say that we have to combine some experimentation out of discipline. That's where I would go.

Now, of course, after we've had this conversation, I know that I'm going to be prioritizing your talk at AIMFra, but are there other things at AIMFra that you're really looking forward to engage you with, whether it's a particular talk or engagement with the industry? Yeah, there are a couple of things that I'm really looking for. I'm looking for talks on like how things like governance, reliability or cause discipline. That has been one of the most asked questions

and how people are experimenting with AI. And more interestingly, very important thing for me is that how people are using consistently AI workload in production. So that is one thing. And how teams are able to build trust with their users or within the organization around AI. That is something that I'm really interested.

interested to learn about as well. Awesome. Rakesh, one final question for you. I know that our viewers and listeners online are going to be keen to connect with you before AI Infra, maybe connect with you about the broader work that your team is doing. Where can they connect with you and find out more? I would say that, you know, easiest way to connect with me is LinkedIn. And if you happen to reach out, just mention AI Infra, so I know the context.

And I really look forward to connecting there and continue. our conversation there. I can't wait to see it in Santa Clara in September, and thank you so much for spending this time with us on Tech Arena. Thank you. Thank you so much for having me here. Thank you. Thanks for joining Tech Arena. Subscribe and engage at our website, Techorina.com. All content is copyright by Tech Arena.