Navigating Data Management Challenges with HCLSoftware’s Kalyan Kumar

Hello and welcome to the latest edition of the Techstrong AI video series. I'm your host Mike, er. Today we're with Kaylin Kumar, chief Product Officer for HT Software.

And while we're talking about what's real and what's not real about ai, Kaylin, welcome to the show. Thank you Maik for having me. Always a pleasure.

I think you're also known as KK to your friends, right? Yep. All right.

Um, walk me through what's going on here. 'cause every other day I will see a story talking about how, you know, AI is not living up to expectations. And then the next story will say something about there's this awesome breakthrough that's occurred.

And it seems like we're on a spectrum and bouncing back and forth, but, you know, you've been around the tech curve a few dozen times. What's your assessment and what's going on here? If you go back and start, look at, you start looking at equating, uh, the whole AI to electricity.

Correct. When real electricity got, uh, discovered, I would say it was never invented because it was always there. Uh, the entire, if you really went through how microgeneration power plants are coming up, ac current to DC current, you had so many different options.

One thing, but more and more people gotted, a lot of new inventions happened, right? People discovered that there are different ways to consume electricity. I think AI is pretty much in the same, you should start to look at it from the same lens.

Uh, if you really go and look underneath the power, what's the real factor around AI is you, is you are able to have a lot of cheap compute. Relatively cheap compute, what you had 15 years, 20 years back. AI has been around for 40 years, correct.

But the ability to process or have cheap compute has been a very, very big factor. And ability to produce significant amount of data and then use the compute to process the data to generate more data. And that's how this whole cycle has been running through.

There is consumer ai, AI in the consumer world, and AI in the enterprise, and AI in the government and in other parts of the business of the whole social landscape, very different. This thing, a lot of the things which you talk about or you only hear about AI, is the consumer side of the ai. It's so much that you've got apple intelligence, you've got Google Gemini on your Android phone.

You've got so many different options. And that's where you are seeing the rapid pivot with open ai, a lot of chat GPT, a lot of those things are happening, or with cloud or misra. Now the latest one, deep seek, if you see this lot being seen on the consumer side, correct, consumer ai, which you and me see it as a consumer of this thing.

And that's where you are seeing a lot of new explosive, uh, uh, innovation happening. And you're seeing that enterprise adoption is, is going to go through its own cycle, but there's a lot of opportunity in there In the enterprise. What are the challenges that people are running into?

'cause you know, when I talk to folks, they're basically trying to figure out how to insert something that's probabilistic into processes that are deterministic. And, you know, it's a, it's as much art as it is science. So I would say we should, before going there, one has to start to really look at what do you need to do AI and enterprise data, which is organized, cataloged, and has the right lineage, has the right tagging.

And very, very important is ability to have very strong metadata management. The definition of data, there is data, but how do you define data? How to say, just so that the definition of data is metadata correct.

And those are the things which enterprises need to really get those foundational models sorted in. And second to your point, which is this whole process, business processes have always, because there are regulations, there are uh, laws, there are constraints. You always have a deterministic process.

You take something, you take it to an outcome. And hence your KPIs, your measure key performance indicators, your process measurement metrics, your cycle time, your weight time, processing time, everything is measured around the steps of the process. But whereas if you really see a lot of the determin or the probabilistic functions or functions are executed by individuals or machines or rules or workflow or applications, which is a human in the machine combination on the fun functions, correct.

Where you're seeing insertion of AI happening is, okay, a gen take or gen AI is around the function, correct? A task, a sequence of tasks. But if really look at one big important thing is everyone is seeing this in silos, correct?

Like every company saying, oh, suddenly you'll suddenly see companies who's making HR software companies who's doing it service management software, suddenly saying, I'm gonna be in the agent tech, I'm gonna govern agent. But they're still seeing this in a very narrow myopic way. What you really need is to really go back and look at what really happened from the time mainframes existed.

They still exist now to how you're really bringing all those new pieces. Your processes are cutting across various different functions and systems. And you need to start to look at how can you or castrate these multiple non-deterministic system to really drive an outcome.

You need to have the right input, output, measurement mechanism to be able to orchestrate the process to an outcome. That's going to be a single biggest focus, which enterprises needs to have. They need to clean their data, they need to really look at orchestrating processes.

And the third thing, which the mic they need to really look at is now how do they build human experiences? Finally, there's human in the loop. I think the way the pendulum is swinging off, like everything say, oh, everything is gonna be done in agent and people are gonna be playing chess, right?

Or, and that's that, that's a pendulum swing. So I think how do you bring human into the loop? I always say these are the three important pieces one needs to start to look at.

So how long will all that take? 'cause I think there's this expectation that somehow or other, this is a light switch that's gonna occur tomorrow when it feels like it might take a couple of more years to get to the outcomes that we're thinking In the consumer world. It's happening now.

There is enough, enough use cases on consumer for you to go and just consume and deploy these capabilities. It's out of, it's easy to attack and consume. There's enough APIs available.

There is a big question on cost models because as you move from zero shot LLM to multi-shot, to rag base, now to tech, you really need a more number of tokens. So if, if you look at the whole play, there is a big cost model around this. The consumer side, if you look at even the likes of open ai, they launched a $200 subscription per month.

Now they're saying they want to rate limit that because they're not able to predict the way this consumption of cost models is gonna be in there. So I would say consumer, it's happening now. It's in there.

A lot of applications being used in, you gotta see this phone, gotta see every other aspect getting a lot of AI embedded. You saw with Apple intelligence, they've taken some of the capabilities from open ai, the backend and put it into this thing. Google is doing that with their Android, uh, ecosystem.

You see Microsoft doing that with the copilot ecosystem, you're gonna see a lot of private implementations of ai. So consumers that sampling enterprise. It's a journey.

I think one very important thing which people have realized, Mike, is that data engineering is a very, very important discipline. But to do good data engineering, you need to, if you really want to do data prep for ai, gen, ai, origin, ai, doesn't matter, whatever type of ai you need to really start to understand your data, your metadata, your data catalog, start to define data products, your data, most of the data is hidden behind applications. You're locked behind, very complex courts applications.

They're locked behind, custom applications. How do you make data having its own identity ability to access those data so that you could now compose and build new applications, new insights, new understanding of this data, which is your enterprise data, your business data. Because you've collected that over the period of 50 or hundred of hundreds of years in many, many corporations, large corporations, they have a lot of data which is getting digitized and, and then they have patents and trends and they could go through and see psychical views of the business.

But that's gonna the single most important thing. There's too much chase around AI and enterprise. The key is get your data strategy right or organize your data, tag your data, and then you need to have a enterprise or guest universal orchestration strategy.

How will all these systems talk to each other? There're gonna be multitudes of agent and you're not going to do an agent with one NLM and one provider is, and or add to that complication. Every SaaS application, every COS application now has an agent tick add-on.

So you gotta have this, uh, army of agents in the enterprise somewhere. You need to organize them, orchestrate them. You need the right, you need to build like the traditional, the way you had the, you might have very distributed model people and then the uncomplicated, this data discussion on data fabric versus data mesh.

You democratize access to data, make it more federated, bring it all centralized control. You do both models. But to do all these things, you need to organize.

All these pieces always go just explore, correct. It's gonna explore and you're gonna have no control over what's happening. So I think enterprises needs to build it around very solid data foundation and then start to look at how do you do enterprise orchestration and then start to figure out how do you build experiences for the users, the enterprise users, the employees, the customers, the partners or the consumers.

Is this a classic example where I think it was Bill Gates who talked about where we tend to overestimate the impact of technology in the short term and underestimated in the long term? Yep. I think he made a very valid point.

I would say, uh, what we have. So if, if we really look back and I think Mike, it's, it's absolutely interesting. Go back and look at five, six years this before covid the same conversation we are having around cloud, this whole acceleration, adoption of cloud.

Oh, there's no more private cloud. Everything goes public cloud, correct. And covid really accelerated that adoption of the cloud, who is actually providing you access to all the major LLM foundational model, the gentech capability, maybe Gentech capability can spread up the core foundational very process.

It's the hyperscalers, it's a new type of workload. If you go back, there was this big chase around migrating infrastructure to the cloud, then they're starting to build new applications on app modernization to the cloud. Then this big push around the data to the cloud was always a big thing.

Not all data move to the cloud, right? So now, and then there was this big player on SAP, uh, how do I now TSAP to the cloud, correct? How do I make SAP adoption and run into the cloud?

Uh, then the rise came and all those pieces came. They look at what's the next big workload is AI is a way for you to move data into the cloud. If you do not bring your data to the cloud or to the hyperscale infrastructure, you can't have access to all this latest and greatest language models and other pieces.

And then the other spectrum is you've got private ai, you can run AI within. That's where now the question is where do you have the GPU capacity available to run those language models and different, uh, yeah, LMS in your data, in your infrastructure now comes this whole deep seek, which again is a big this. They say, okay, they use far less at compute, they use more CPU processing, then you have AI PCs, then this whole conversation on small language models.

So what's, what's really happening if you scratch everything off the surface? Majority of the processing capacity to be able to use this language models exist in the hyperscalers for you to consume that, you need to bring your data into this. And then you have this enterprise applications which are still spread out in SaaS, in in package damps, within the data center on the cloud, all those different scenarios.

So if you, if you sum all those pieces together, this is a journey which an enterprise has to move. But in the hindsight, it's the right thing. Because if you're dealing data and applications, because the values in the data app is how you interact and experience, it's also gonna be a very, very interesting moment of how do you, would you re-look at your app modernization strategy or app upgrade strategy?

You'll say, why the hell do I do spend money on modernizing applications if I can just use different methods and technique to access the same data in a conversational, uh, context or make it consumer a rich experience context. Why do I need to worry about spending money on modernizing the application? Let me extract the value out of the data.

And, and now with processes and orchestration, other capabilities, you could could do that. So it's a very inter, we are in a very interesting period, pragmatic decision making is going to really, but customers are gonna step back and think seriously saying that, where do I need to spend the money in the right way? So what is your best advice to organizations then about how to go after this?

Because I think a lot of them are struggling with data scientists and they need data engineers and then they gotta bring developers in a couple of security folks and um, before too long it takes a village to get anything done. So how do I organize this thing? I think the enterprises first thing first, they need to really step back and do three very fundamental pieces things, right?

You have your current IT systems landscape that continues. What doesn't stop on what you're doing today as part part of a digital transformation journey. Very, very important aspect is one is step back and look at your entire data strategy.

Look at data, develop a holistic view of how do there is, forget the AI piece. Get your data strategy right? How do I organize, tag, understand the lineage, create enterprise data catalogs.

Some cases we want to create data products or data contracts or data. Just there are different mechanism techniques. How do you really understand your data and hence have a very strong mechanism of doing active metadata management, understanding your data and how do you manage the data itself?

The definition of data where it exists and it's gonna, you cannot control the origination processing, post-processing creation of new data. It's gonna happen across the enterprise, right? It's gonna happen because we have so much digital technology around and data, right as we speak petabytes of data is getting created or gigabytes or terabytes of data is getting created within the enterprise.

So the first is look at data strategy, which means if you step back and look at how do I understand what data landscape, you'll get a lot of aha moments. Oh wow, I have 17 copies of the same damn data called with different names in different systems. Great, you don't wanna go and clean all of them on day one, but now you can join all of them and say, okay, this, let's take a simple example.

Employee ID could appear as M code, E code MP or code employee underscore code employee code. It could appear in 58 different systems. But now you need to understand, oh, where it exists, which systems are using it, who is the consumer of this data?

Which applications, who is feeding and where is it coming from? Where is it going? Understanding the lineage.

Once you know your data prep, you understand you can annotate tag, you have to democratize access to data. You cannot do everything in a fabric model. You need to have critical stuff in the fabric model, okay?

Uh, and then, but you'll have to have a more data mesh approach. Correct? Distributed, it has data, has to be every business function.

Employees will have access to the data based on how you manage the data access. That's the first thing. Second, enterprise have to actively look at an enterprise orchestration strategy of how do I connect mainframes, distributed, ERP, SaaS on-prem, all the systems because the processes will run in the systems.

But can you look at, uh, even things like techniques which people used. If you go back when we had batch processing workload management, we had dialogue processing all those systems which existed, credit processes always ran and batch long, long running processes, interactive processes, correct? That's how it really happened.

Things of techno. But, but understanding those pieces, understanding where they're running today, developing a holistic orchestrator process orchestration, data orchestration, and that whole piece. Now a piece of that orchestration is going to be AI orchestration.

What's gonna happen in the AI origin AI agent, right? But you still need to orchestrate a lot of other things. Like if you wanna do money transfer in the bank, it's highly deterministic process to take money from A and move it to B.

You might have a lot of AI to do a ML fraud detection, this and that, but money moves away to B. It's a very well orchestrated process. How do you make sure these different systems are executing the process?

You've got the right enterprise orchestrator, which can watch the process execution, understand the in and out of those pieces. So hence universal orchestration. One is active metadata management, looking at how you really understand your data.

Second is process orchestration strategy. The third, how do you build new experiences for your enterprise employees, consumers, businesses, partners, customers. So there's human in the loop all, you can have a lot of AI in this.

But the third, because hence I would put it in a very simple way of XDNO experience data and operations and operations. You orchestrate using a very enterprise orchestrator or whatever. If you have IT operations, you orchestrate IT operations, it's fine.

But we you, how do you look at operations orchestration holistically? How do you understand your data and how do you make sure you create data products and data catalogs so that people can consume and occur? Is how do you build your experiences now in this mix that will be so many, you'll have too much choices, right?

Every package app, which you have bought will now have an HHL ticket add-on. Take it to me in writing. If you don't have today in the next six to nine months, everyone is going to give you.

Now you need to make those choices. And that can only happen if you understand your data and your enterprise orchestration and your human experiences, then you can make a choice. Hey, ASEC is built a very nice agent for finance.

Ah, let me take it and just plug it to my orchestrator and say, deliver it straight to the human experience. Hey, someone else has built an agent with 70% functionality. Can I build an experience orchestration in the front end so that I could just add on those pieces for the user to do that?

That will only happen once you do this three all is what's gonna happen. Everyone's going to give you their own system, their own processes, and now you are gonna be trying to figure out how do I stitch all of them together to deliver experiences back to the same square one situation. So I would say this is the right time for enterprises to pause.

Look at a lot of these pieces together, Folks. You heard it here for all the algorithms in the world still comes back to how good is the data. Hey kk, thanks for being on the show.

So same here. Thanks Mike. And I hope, uh, I could be of use to the listeners and the viewers.

Uh, uh, one small step. Keep educating our customers in terms of what I hear from others and help take a pragmatic step forward. And thanks for having me.

All right, folks. Thanks for watching. You can find this episode and others on our website.

We invite you to check them all out. Until then, we'll see you next time.