What we talked about
Nithin Thekkupadam Narayanan joins the show to unpack the craft of product management at scale:how to keep a mature platform stable while innovating fast. A Senior Principal PM (ex-AWS, ex-Remitly and an MIT SDM alum), Nithin shares pragmatic lessons from event-driven systems, cost/speed trade-offs, and the fast-moving AI landscape.
Show notes
Nithin Thekkupadam Narayanan tells a story that most product managers would rather not admit: he launched a feature that would save customers 40% on storage costs, and almost nobody adopted it. The lesson he took away is that cost savings are not always the king, and that understanding the real priorities of mission-critical users is far harder than reading the market signals.
What we covered
- Event-driven architectures are becoming the backbone of AI workloads because old data is stale data. Nithin’s product, Oracle’s Transactional Event Queue, evolved from writing to disk (slower) to writing directly to RAM (faster), and then to natively supporting vector data types so that AI workloads do not lose time on format conversions before accessing data.
- The speed-cost tradeoff is not theoretical. Nithin described a small company that ran performance tests on Kafka, forgot to shut down the infrastructure, and received an end-of-month bill that exceeded their entire annual IT spend. Every design decision about speed involves a corresponding decision about what you are willing to pay for it.
- Launching a managed Kafka storage feature that cut costs by 40% produced unexpectedly slow adoption. The reason: when you are running mission-critical workloads, “if it’s not broken, don’t touch it” takes priority over saving money. A year after launch, many customers who said it sounded great still had not activated it.
- On feature prioritization, Nithin’s most consistent method is quantifying the signal: bucket similar feature requests from different customers into themes, count how many customers share the underlying need, and use that data to make the case to internal stakeholders. The biggest failure mode is being in limbo, a decision that is neither yes nor no is worse for the customer than a firm no with a clear reason.
- Oracle’s Converged Database consolidates relational, document (JSON), graph, and vector data types into a single system, with the Transactional Event Queue built directly in. This eliminates the need to manage a separate message broker alongside multiple specialized databases, reducing an architecture that might require five or six systems for a single use case down to one.
- The EVP at Oracle framed it this way: “AI is not going to replace people, but it will replace people who do not know AI.” Nithin described using deep-research AI workflows, not just Q&A with an LLM, to synthesize market data and build the internal case for new product investments, compressing a process that would previously have taken weeks.
About Nithin
Nithin Thekkupadam Narayanan is a Senior Principal Product Manager at Oracle, where he leads the Transactional Event Queue product. He previously held product roles at AWS and Remitly, and holds a degree from MIT’s System Design and Management program.
- LinkedIn: http://www.linkedin.com/in/nithin-thekkupadam-narayanan-0bbb4384
- Website: https://oracle.com
Episode 78 of the PreVetted Podcast.
Full transcript
Federico Ramallo (00:00) Welcome back to the pre-vetted podcast where we spotlight extraordinary people and remarkable talent reshaping our world. Today we’re joined by Nitin Thekupadam Narayanan. He’s a senior principal product manager at Oracle. Formerly he was at AWS and before that he was at Remitly. And he’s an MIT SDM alum.
So his work span from event-driven systems, fintech and research on improving scaled agile with design structure matrix. So we’re going to dig into Oracle transactional event queues, TX event queue, real-world customer stories and the craft of technical product management. Nathan, welcome to the show.
Nithin (00:45) Thank you, thanks so much, Frederico. It’s nice to be here.
Federico Ramallo (00:48) Yes, we’re honored to have you today.
how do you balance having a well-established product with the need to innovate quickly?
Nithin (00:56) That’s great question. Something that PMs, most of them would agree that they have to deal with on a near daily basis. So I’m with Oracle that has been around for eons. feels like it started out in the seventies. We have extremely, extremely stable products that customers have trusted for ages, but
It never ends there, right? It’s technology used to change earlier per decades by the year. Now it’s months. It’s from inception to maturity. It’s moving within months. So, know, latest case in point, if you think about the AI developments and not talking about the LLMs, we all have seen how they have evolved. But when you’re looking at things like model context protocols,
or the latest protocols that are coming out. Protocols generally take a lot of time to mature, but MCPs, for instance, they were released by Anthropic in November, 2024, and it already reached mass adoption within a matter of a few months. And already other companies are building newer protocols for agent-to-agent communication. So as a PM, the first thing is…
listening to your customers, what are their needs? What do they want to improve your product? And always keep an eye out on the market and what’s coming up new. You may not always hear from your customers and exactly what they need. That may not be the step in the right direction. Then you are building a product for one customer. So it’s important to balance out what you’re hearing from your own customers and what’s happening in the market to make the
best judgment. that would be an overarching statement that I would make. And there are various principles that PMs use like ranking your requirements and stuff like that. So I’m not going to get into the technicalities of that. But for me, in my personal experience, has been in recently, I’ve been working with a lot of customers and the product that I work on, its predecessor has at least been there for 20 years. And to cater to the customer’s need for
processing data in near real time. We have made a lot of changes in our product to ensure that the data can move faster. So if you don’t mind, I’ll go slightly deeper into the technicalities of it. yeah, appreciate that. Yeah, so I’m working on a product called transact 11 queues and
Federico Ramallo (03:16) Of course, of course.
Nithin (03:22) queuing systems and messaging systems are increasingly becoming, or even streaming in general, is increasingly becoming the backbone of AI architectures. Because it’s no more that you have some data, you wait to process it, and then start making decisions on that. In order to train the models or even make real-time decisions, you need access to that data in near real time.
My product was previously based on actual hard disks. So you get some data from an outside into the database through the queues. It writes onto the disk. Then from there you read. So that slows the whole process down. So we improve that based on customer demand so that the data actually gets written only onto the memory. So RAM is always faster than your actual disks. Right. So we access the data directly from
the memory so that customers can access this data quicker. And when I say event-driven, that means it is very reactive. So that’s how customers expect the business to run. So for example, if you’re using a credit card, somebody else has a fraudulent activity is happening on your credit card. You cannot wait till that.
the fraudulent activity has completed a couple of days later you get to know that my god my credit card has been subjective. Then you have to go through the whole conundrum of disputing the whole thing and it’s trust me I’ve been through that it’s not fun. So now if you look at your credit cards even if you are accessing your credit card from a different country while you’re traveling you get an immediate notification if you said hey is this really you who are making this transaction. So this all is
basically event-driven. So you swipe the card and event is generated that, your card has been swiped. And in near real time, you receive notification, your servers receive notification that something has happened and it goes on to the client, basically your phone saying that something has happened. unless this happens the very second or within milliseconds, to be honest, you’re losing the value of that particular use case. So old data is stale data.
So this is where products like transactional event queue help customers to ensure that you’re getting the data as fast as it is being produced and being able to process it in a very similar fashion. now going back to your question, right? How do we balance all this? So we have evolved from using a display system to an in-memory system. And now when it comes to agentic AIs, which is, I would be remiss if I don’t mention that.
Everyone is trying to build agent-kai and it comes to the, it presents a lot of problems and challenges because the languages that customers are using are changing.
the data types that the customers are using are changing because earlier it used to be an RDBMS database. Now the data types are changing, JSON data types, document data types, graph, vector. Vector is very important for AI workload. So all of these things are changing and the demand of the customers are changing. So as a PM, I need to evaluate where things are going next. So for example, recently while AI is making it in root, how can we help those customers?
The answer to that is start natively supporting the data types that those customers would want. So we made our changes to ensure that my product, so if the data is coming in form of a vector, we don’t have to do any conversions or anything. We natively support and pass through so that customers don’t lose time in accessing that data because some additional processing is happening. So this is how I feel even if it is an established product, if…
they do not evolve with the evolving needs of the customer and the market, it would be redundant. I hope I have answered your question. Yeah.
Federico Ramallo (06:59) Right,
right. Wow, so much to digest. I agree with you that having a reactive system scales better. I used to build these pull systems where you have a, I mean.
Nithin (07:04) I know, yeah, I’ve been learning it the hard way as well.
⁓
Federico Ramallo (07:15) If you increase the frequency and you have many clients, then everybody’s pulling and pulling and pulling, then you have a bottleneck, right? Or you start using more resources just to pull, just to ask for queries, right?
Nithin (07:25) Ooi.
Yeah, you brought up a very, very important and interesting point. It uses a lot of resources. So when you move at a large pace, that is something that we as providers and customers also have to consider. Yeah, you need things at speed, but at what cost, right? So especially in a cloud model when you are paying as you go, if you…
have use cases which do not need that amount of speed. let’s face it, it’s not that every single use case has to be done in milliseconds. Few things can wait. Right? So it’s always important that you think about the trade-offs that you have to make when you’re building these systems. So some use cases need the speed, some use cases don’t need. And more often than not, when the speed matters, there is a correlation between the speed and the cost that you’re paying for that.
So I’ve had a customer in my past. I’m not going to name them or the company that I was working for. So they were trying to test out this feature using Kafka, which is one of the most popular messaging platforms or even streaming systems. And they were doing some performance tests and they forgot to switch off that system or delete that infrastructure after the testing. their end of the month bill was more than their overall IT spent.
This was a small company. So that’s how much it can bite you if you do not pay close attention to how much you are utilizing. So yeah, I appreciate you bringing that point of how resource consumptions happen when you work on a model like this.
Federico Ramallo (08:43) Wow.
Yeah, just the HTTP requests has so much overhead, right? That just, if you’re pulling and asking, is there any news, then the amount of bandwidth that you use, amount of CPU resources that you use for doing basically nothing, right? Because you’re just asking for updates. It’s huge, right? I mean, I was able to improve that by having this long, long…
Nithin (09:13) here.
Federico Ramallo (09:18) connections. So we have these low-low connections and then we will ask frequently. But as you mentioned, there’s still a cost associated with that because now you have to keep those connections running, right? And the server has to support it. So we tried to work around with that, which kind of worked, right? But that…
Nithin (09:19) Yeah, long-reeling connections here.
Federico Ramallo (09:37) But we try to do that very specifically, for very specific use cases. When you’re trying to prioritize interactivity over performance. But then when you are doing notifications that you know that the user can wait a few seconds, then you have so many other options that you can do that.
Nithin (09:54) Yeah.
Federico Ramallo (09:56) Yeah, and then that scales up to a point when the bandwidth becomes an issue, the CPU resources become an issue, right? And that’s where the messaging system helps because now you have a system that is specialized on that that basically just works as a kind of broker. Maybe broker is not the right word. It’s more of a…
Nithin (10:14) Yeah, it is, it is actually, yeah.
Federico Ramallo (10:18) really, okay. I
was thinking more of an orchestrator or coordinator, but yeah. And basically, can do that in a much more, you know, it can scale much faster, right? And then you have different people connecting to different channels, expecting different news, right? As a way to kind of filter what you’re talking about, right?
Yeah, and then the issue that I found there is, and you’re probably familiar with this, is when you’re broadcasting news, right, you have to be careful about making sure that what you’re broadcasting, the people that are listening to that broadcast, you know, there are no security leaks into that because you could be broadcasting an information that is, you know, sensitive, right?
Nithin (10:57) yeah,
yeah, absolutely. exactly, right? This is another very important trade off that you always have to keep in mind, security. So what are the level of access if it is a need to know basis or is it a public? And it’s not just about who’s accessing it, where we are hosting it. Is it something that is publicly available on the cloud or is it a private VPN? And a lot of customers still prefer hosting it in their data centers.
for these very security needs. yeah, and which is precisely another point, right? So when we are thinking about a product and a new feature, we need to ensure that it’s available at the right places based on the customer demand. So just because cloud is popular, it doesn’t necessarily mean that we have to always build our feature. It becomes easy when it is cloud, it’s easier to deploy features into cloud.
But a lot of customers still see the value of on-premise servers. So we need to ensure that we cater to them as well.
Federico Ramallo (12:00) Yes, yes. I think there’s a lot of value on using the cloud and cloud providers. But there is a fixed cost when you’re implementing your own servers. There’s a fixed cost on…
purchase, maintenance, right? I mean, now it’s easier because it used to be that you had to buy the physical hardware, put it in a physical building, right? I used to do those type of jobs, setting up those racks and servers, and it was interesting. ⁓ But then, know…
Nithin (12:16) See ya.
Okay. Yeah. Cool. Yeah.
Federico Ramallo (12:30) the first version of the cloud was we can give you a virtual computer, right? Or a physical computer, but you don’t have to set it up, right? So that was huge, right? ⁓ And then they start adding more specializations on other services, right? But there’s a fixed cost to having that. And then you’re committing to a bandwidth or a…
Nithin (12:38) Right. Yeah. Okay.
Federico Ramallo (12:49) a capacity that if you don’t use it, you’re still paying for it. ⁓ So that’s where I see the value of the cloud because you can be flexible based on the demand, you can ask for more resources. And if you don’t need them, then you can reduce the resource consumption. And then the other thing I saw with some companies where they change the APIs very frequently. ⁓
Nithin (12:53) Yeah.
Absolutely, yeah.
Mmm.
Federico Ramallo (13:14) Basically, that means that when you’re the application, it breaks because what you build now doesn’t work. I understand that they are trying to innovate quickly, but then on the other hand, if they support the different versions of the API, then the amount of…
Nithin (13:19) Yes. Yeah.
Federico Ramallo (13:35) complexity grows exponentially for them. So I understand from their perspective trying to innovate but not keeping legacy or API available. So I can understand that, but on my side when I’m consuming those APIs, it becomes frustrating.
Nithin (13:51) Yeah,
no, no, very true. Yeah, that’s a, that’s a constant challenge. So a lot of companies I’ve seen, they support at least few versions, few older versions of that API as well. So that it gives customers enough time. But that adds cost to the company because you have to again be on a maintenance mode for all that. So if you’re making any changes, you have to ensure that it, it is backported to all the older versions so that it remains compatible. But at some point they have to say, okay.
We are done with this. We have to stop supporting. But it’s important that you give customers enough heads up before you retire anything. Retiring stuff is as important as launching new stuff.
Federico Ramallo (14:24) Right.
Right, yeah. So what is the most surprising insight you have uncovered from customer feedback?
Nithin (14:37) it was a few years back when I was launching one of a feature on, you know, a managed service for Kafka. It was a big cost saving measures. And this was right in the middle of pandemic where, you know,
growth of pandemic had, know, a lot of companies expanded exponentially during the pandemic seeing the opportunities and that bubble was starting to or rather that that growth was starting to flatten out and we launched a feature that was extremely cost effective and this was for Kafka.
managed service for Kafka and it was a storage service which would reduce the cost by easily 40%. And there was a lot of noise in the community for a couple of years and we decided, know what, Kafka is open source, right? So we are at the mercy of community for new features to come. And that firm, we had to make a decision that, you know what, we cannot wait for the community because it looks like the customers might need that sooner.
So let’s go proprietary and launch it for the customers and then see whether the community will follow suit. While it did, but when I saw that, you know, cost it’s, you know, the market is going that direction. Cost is being top consideration. That’s what the trends are showing. So if you launch a feature like this, people would just, you know, come in hoards and start adopting this feature. But to my surprise, it was a very, very slow adoption.
So it was more like customers were leaving money on the table because you could easily activate it. what I realized was when you’re powering mission critical workloads, Kafka is always very mission critical. So customers, if it is not broken, let’s not do anything. So that is the priority and not just purely cost. my big learning was cost savings may not always be
the king or the first top of mind, it is peace of mind for customers. yeah, especially when it comes to hypercritical product, you know, very critical and complex products. There is, there is always a, you always have to think twice before you deliver it and really, really have to talk to your customers, possibly how many ever you can. It saw great at options.
because there were different categories of customers who are savvy enough to think about it, but there are other customers who are slow to pick up. So just because launch something that would save a lot of cost for the customers, it doesn’t necessarily mean that you will see magical adoptions. that was a good learning for me. know, dollars are important aspect, but you need to look at the usability and
the actual priorities of the customers at that given moment and go to market accordingly.
Federico Ramallo (17:27) Interesting, interesting. I mean, it’s the issue of sunken cost, right? you know, people prefer to pay more rather than the potential, you know, cost, whether it’s emotional or money, right? Yeah.
Nithin (17:32) deal.
Yeah, yeah, and
that emotional and it’s a lot of work at the end of the day to make changes ensure that their mission criticals there is nothing going wrong. There is always a skepticism when something new comes in right, especially on a system that is already working okay. So yeah, got a lot of customers who I spoke to yeah, this is great. We’ll think about it and
A year went by, they still haven’t thought about it because things are working, let’s not prioritize it. yeah, that’s an important trade off that you have to think about.
Federico Ramallo (18:14) If it’s not broken, why change it? Why fix it? But then on the other hand, saving money eventually becomes a priority and then that’s when they…
Nithin (18:17) Fix it. Yeah.
Yeah. So when we were going to market,
I was focusing on customers who actually, you know, would value a cost because there were a lot of customers who were, you know, doing a lot of cost cutting. So it’s not just about striking that opportunity, but thinking from that customer standpoint, because the thing about these features are just because you launched a feature, doesn’t mean that everyone would know about it. There’s overload of information out there.
So you need to ensure that you go to market and go to the customers who actually need this. and use all the mechanisms at your disposal to reach out to those customers, be it through marketing or be it through partners and so on. So, we need to ensure that our go-to-market strategies are actually reaching our right target customers. Just reaching to everyone saying that, this is going to save costs for you.
is not the best use of their time or yours.
Federico Ramallo (19:14) it doesn’t resonate with them.
The other complexity with that on the go-to market is the complexity, right? Because, I mean, now we’re talking about it, and in retrospect everything, all the decisions that you made, it was obvious, right? But at that time, you don’t have all the information, right? And there’s so many moving parts. By the time you do something and then you get the feedback of whether it’s working or not, and make the corrections, right?
Nithin (19:34) yeah.
Very true.
Federico Ramallo (19:50) it makes it harder and harder to figure out what to do next. And if you don’t do something, if you don’t do anything, if you don’t do anything, you just freeze. Then you’re missing the opportunity to do something, right? So there is, know, them if you do, them if you don’t, right?
Nithin (20:02) Yeah.
It was a very humbling experience for me also, a big learning opportunity there.
Federico Ramallo (20:11) So how do you prioritize features when different teams have, all have compelling demands for the same product?
Nithin (20:18) Yeah, been in the industry for 20 years and I’m still learning. This is kind of the hardest problem I feel. These priorities competing. if it is… First thing, first what I’ve learned is you cannot please everyone. You cannot please even your entire customers. Right? And I do not have a straight answer for this. I guess the answer would be it depends.
Right?
There are tons of driving factors and a lot of people have tried to put frameworks around it. And in my personal opinion, at least I’ve seen few things that work. what I generally do was look at feature requests that are coming in. I’ll tell you a few mechanisms, but these are not foolproof or these are some things that I have seen helping me in the past, but not necessarily would work all the time. So for example, when you’re having feature requests, so this is something that
are customer initiated. So this directly gives you some voice of the customer. So if they, if they tend to see that there are some features that they require and periodically I check, okay, this particular feature seems like has more demand. So you can quantify that based on what customers ask us. They may not mention the exact same features, but you have to drill down a couple and try to bucket them and figure out, okay.
Maybe these guys need some ways to improve their performance and we have this to offer to do that. So let’s start supporting that. So for example, in my previous job while at Amazon customers really wanted to, again, this is where the cost was king. Customers wanted to change their processes or start using more cost effective.
CPUs and Amazon already had something like something called Graviton Which is a cheaper alternative to few other processors out there So customers were like hey, why don’t your service also start supporting that so we can reduce our overall spend? So when we look at those Voices coming in form of the product feature request. This was one of the top ones So that is one way in which we can quantify when your product has good adoption you have
you know, thousands of customers already, then it is easy to figure out what next to build. Not easy, but at least you will get some feelers on what customers are looking for. Another effective thing which I generally do is roadmap transparency meetings with customers. So customers who are fairly engaged partners or customers.
We do share the roadmap with them, have conversations on a frequent basis, telling them that these are the things that we are planning to build. What are your thoughts? Does this resonate with your needs or are you going in an entirely different trajectory? So getting those sense from the customers makes a lot of sense. And then comes the difficult part.
Right? The actual building. So like you said, right? There are different stakeholders in the company and product is not like just a team of folks sitting and building something. There is a ton of integrations. Like you said, APIs may have changed in another team, which may have dependency on the feature that you’re trying to build. How do you convince them saying that, we need to make this change because X customers. So this become an, in a lot of job descriptions for PNs, it’s easy to see that you need to be able to influence decisions. And this is precisely what that is.
So you need to be able to build a case for why this feature is important, not only for this product and perhaps for the organization itself, because these may be things that impact your bottom line or the top line, right? So that would be the next steps to make those decisions, analyze the pros and cons, bring people together who would be, who you would need to partner with them.
partner with in order to deliver those products or features and ensure that you are getting a consensus. And if it is a no, for me the biggest challenge is being in a limbo. It’s decision has not been made but it’s just getting prolonged. That is frustrating for anyone and more so for the customer. So if it is a no, it’s easier as if the stakeholders say, we do not have this, we have
bigger fishes to fry and we cannot deliver this this year. Right? But if there is a compelling reason why which they’re saying that despite going to the single threaded owner of that particular org, we have a definitive answer. So having that definitive answer, yeah, it may not be pleasing to the customers, but at least you can say that, this is not something that we can do at this point, but we have definitely taken it into consideration. So, and one important thing that I’ve learned is
Don’t commit until we have a definitive answer like this. Just go to the customer and say, yeah, we’ll do it for you. And six months later, you come and say that, hey, sorry, I’m getting pushed back. So something that I have, I always ensure is that not to, it’s not just CYA. It is also about setting the right expectations with the customer. I feel that in the long run, customers respect that, than opposed to somebody trying to, you know, make them happy by saying that, hey, we are going to do this for you.
Federico Ramallo (25:23) Right, if you say yes to everything, then you’re over committing and then you have the engineering team working overtime to deliver. And then when you deliver, the customer says, well, that’s not really what I wanted. So it’s challenging, right? I think it’s a balance between sensitivity, but also being able to push your vision, as you mentioned, influence on the decisions.
Nithin (25:24) Yeah, exactly. Yeah.
Yeah, that’s not good.
Yeah.
Federico Ramallo (25:49) because
Nithin (25:49) And at the same time, the right expectations with the customer. We’ll try. And always important to understand whether it is a must have. Is it something that would impact the customer’s business directly? Or is it something that they’re looking to improve on? So all these things needs to be factored in into those decision making when you’re doing the priorities. Yeah.
Federico Ramallo (26:10) Right, right. And the other parties, customers don’t necessarily know what they want or what they don’t know, right? So it’s a little bit of explaining the vision or the reason for the feature, right? But also listening because you need the information from them as well. Yeah.
Nithin (26:28) Yeah.
Federico Ramallo (26:29) When you were talking about prioritization, remember this. I used to work on, I’ve done many, many different roles, but one was IT support at one time. And I realized that we were doing this prioritization by who is the loudest, right?
Nithin (26:40) Thank you.
May the loudest voice in the room win.
Federico Ramallo (26:49) Which is a
really bad way to prioritize, right? But there’s always that user that comes to the IT support office and explain that they have a problem. And they basically jump to all the software that we have for prioritization. And they just get prioritized faster just because the person is there, right?
Nithin (26:56) Absolutely.
Federico Ramallo (27:17) Yeah, and usually something small, right?
Nithin (27:18) Yeah. This is why I like a lot of data driven decision making. Like I told you earlier, right? If I can quantify the justification, that is, that is, that is what I have seen as, you know, putting your best foot forward. So if you have enough data to back your claim on why this
change or new feature is required or the new product needs to be launched, that becomes the single version of truth as opposed to the decibels you have in the meeting room.
Federico Ramallo (27:50) Right, right, yeah. with data-driven decisions, you can make better decisions overall.
Nithin (27:51) Yeah.
Peace.
Yeah,
it has the highest likelihood of succeeding.
Federico Ramallo (28:01) Yes, yes.
So how are AI cloud or automation affecting customers in ways they might not expect?
Nithin (28:10) How can I put this without being controversial is what I’m thinking. There are multiple schools of thoughts, but let me put this from…
the perspective that I have seen with my customers, right? or first of all, I personally see a lot of value in AI because the way in which I’m being able to do the research and reduce that cycle time, it’s not just noticeable, it is significant. I’m right now working on a market research for a new product, which I cannot unfortunately reveal at this point. And…
My ability to do that research in order to help me convince whether this is first, need to convince myself whether this is a good value prop or not. And then my F gone as far as convincing my VP that this is the good way to the next best thing that we can invest in.
But with LLMs and all that, it’s not just about Q &A with an LLM. It’s about that deep research that I’m being able to do and connect a lot of several dots and synthesize the research with both hypotheticals and actual data as really improve the productivity. recently our EVP, Loiza, at Oracle, his statement was,
You know, AI is not going to replace people, but it will replace people who do not know AI or who are not savvy with that. So that’s where it has come to. So use it to the fullest to make yourself the smartest version of you. Otherwise, you know, it may soon make people redundant. So that is what I see from an AI. My personal view, that that statement actually resonated very well with me, which is why.
Yeah, you know, that’s the first thing that came to my mind when you asked me this question. And when I’m looking at my customers, there is a lot of investment and a lot of excitement that I’m seeing and I’m seeing value as well. So, in which customers, but yeah, just because there is AI, doesn’t mean, it’s resolving every single thing. So being intentional about AI, that’s what I, and the downside of this is I’m.
Seeing a lot of
companies doing this wrap an LLM in a nice presentable way and just selling it as a smart bot or an agent that that doesn’t solve too much purpose because it’s just it’s cost because all these things need to run on really expensive GPUs, which is the hottest commodity in the market. Right. So you have to be, you know, customers have to be really intentional about how they use their AI.
Again, otherwise this is going to give you a very unpleasant bill at the end of the month. So, and when I’m working with the customers, from my product…
We focus on enabling the customers to run AI, not telling them how to build AI. So for example, taking a step back, there are a few interesting things that you touched upon earlier, and I’d like to bring that all together and how my product helps it. Sorry, not making a product pitch, ensuring that we need to think holistically about this. So you spoke about security, you spoke about
effectiveness of having real time and utilizing the resources wisely without compromising too much on cost. Right? So my product, the transactional event queues, it’s a built-in event streaming platform within the Oracle database. So what I mean by that is a lot of times, you know, customers have a messaging platform, like the broker that you mentioned, which moves data.
and then you bring it into a database for further processing. So you have two systems to manage.
So what a built-in event streaming solution does is you have only one database and this is a feature within the database. So you do not have to manage multiple products at the same time because managing a message broker is a full-time job for a large team. And which is why a lot of managed services have become extremely popular because you know, services like Kafka, it’s highly configurable, which is extremely beautiful, but more options that you have
the more confusion it becomes. So whenever something fails, it’s a whack-a-mole game. Hey, where did it fail? You press this, okay, something else blows up. So, customers were increasingly frustrated with that, which is why it started becoming a managed service where com, know, companies start like Oracle, AWS, Confluent, lot of them started offering managed service for it to reduce that. But that managed service may not always take away the added cost that comes with it. So when you have a converse, so Oracle has something really beautiful.
called a Converse database. And the value of that is you have, like I said, right, there are a lot of data types that have evolved. So initially there was only RDBMS, then along came DocumentDBs, which handles the entire JSON format. Then came graphs, Facebook and all uses that a lot. Then came vectors for AI. So for all this, there are different databases.
So at end of the day, when you look at it, the number of databases that you’re having to manage is just growing. And then you have a message broker on top of that. So if you need one use case to integrate this, you have some five, six systems that you pull data from for a single use case, which serves few people, mean, few customers, but it becomes a, you know, operational nightmare for a lot of others. So having converse database, so Oracle’s converse database,
is all these data types are supported within a single database. So it helps maintaining consistency of the data, which is of paramount importance to customers, especially in the healthcare, financial services and all that. They cannot afford to have poor data quality. It not only affects their end customers, but also brings a lot of challenge when it comes to compliance. So having everything in one database,
helps these customers. And on top of that, we are having the messaging system or the queueing system also built into this database. So you’re having an entire consistency with transactional capabilities. When I say transactional capability, right? What it means is, so for example, a lot of data is coming into systems like Kafka. Something fails. A lot of times you don’t know how much data has actually come in. What is the last piece of data that came in?
Then you have to trace back the entire path to figure out, what has actually come in? It’s just like walking out of a grocery store, your bag tears and something is left. You don’t know what all fell out, what is there. Then you have to go all over your parking lot to figure out, what went through, right? But when you have transactional capabilities, right? It gives you predictability on how much data has actually been received versus not. So.
Yeah, you may not have received the entire data, but at least you know what data you have not received. So having this transactional capability built into your messaging system and your database for all data types within a single unified infrastructure reduces a lot of complexity that arises from this fast moving data that we have today. So that is the benefit of a converse database and having a transactional queues or messaging system built into the database.
But one could argue that, okay, if this goes south, everything goes along with that, right? So we have disaster recovery systems and all that for it to take care of it. But of course, like any other system, is two schools of thought or multiple where some customers find value in this, others find. So, but this is an alternate architectural pattern that I have seen that customers are increasingly adopting and liking. Think of it as a smartphone, right?
It’s no more that you have to carry 10 different devices or 10 different hardware for your purpose. It’s just that everything is in a smartphone. But still there is market for specialized cameras, specialized headphones, and all of that. So yeah, there is market for everything. But here in this architectural pattern, I’m seeing increasingly
Increasingly, customers seeing the value in something that is converged. So we’ll see how the market trends in the future, but at least now I see this is our best foot forward and this is the right thing to do for our customers.
Federico Ramallo (36:35) Right, You were talking about Kafka having too many parameters. That makes it complex. I love open source projects. I think Kafka is supported by Apache, right? And it’s great, but then I had the same issue. I started with Ubuntu, the first version of Ubuntu where you had to compile the kernel.
And it was leaps forward compared to other distros. But still, eventually it had so many options that you could spend days just setting up the operating system. And it had the same issue with other software tools. When you go to a commercial product, then you have…
Less is more, Basically what you want is to set it up and make it work with smart defaults. And then if you have to tweak it, then you can, right? But, yeah, it reminds me to our bodies, right? Like I’m getting 40 now, so now I’m having a…
Nithin (37:14) Mmm.
Exactly, yeah.
Federico Ramallo (37:30) pain on places that I hadn’t had before. So it’s like, there is a muscle there. There is a joint there that I never knew existed, right?
Nithin (37:38) Yeah, actually I
hit 40 last year and I was not ready to accept it and I went to the gym and the first deadlift I hurt my back and I was down for two months. Yeah, that was a rude awakening.
Federico Ramallo (37:53) Yeah, yeah, it’s hard. It’s kind of the body saying, from now on, I’m going downhill, right? Yeah. I take my son’s school on the motorcycle, and we fell off a few months ago. I remember exactly when. ⁓ Yeah, was part.
Nithin (38:00) Don’t say that! Kidding.
Who thinks? Who?
I’m so sorry.
Federico Ramallo (38:17) part of the adventure of riding motorcycles. There are two people on motorcycles, right? Those that fell off and those that are going to fall, right? So I got my shoulder hurt and I’m still recuperating, but my son was like, boom, up. Let’s go. I’m running late for school and I’m like, wait a minute, give me a second, I have to recuperate. Yeah, yeah, yeah. So.
Nithin (38:24) Hehehehehe
Great. Yeah.
Federico Ramallo (38:41) It’s not good, yeah.
Nithin (38:44) Yeah,
I’m realizing that, yeah.
Federico Ramallo (38:47) Yeah, and I’m training frequently, but still, know, the recovery time gets longer and longer.
Nithin (38:53) Yeah,
and the number of supplements that you take increases.
Federico Ramallo (38:57) Yes, yes, more and more pills, right? Yeah. So anyway, sorry for the discretion, but going.
Nithin (39:00) Yes.
No, no, that was a good anecdote, yeah,
and a good reminder as well.
Federico Ramallo (39:09) Yes, yes,
by the way the eyesight as well, I start having eyes issues and yeah.
Nithin (39:15) man,
we should talk more.
Federico Ramallo (39:16) Yeah, it’s not bad. It’s kind of OK. People tell me, you should expect at 40, you’re going to start needing glasses. But I was like, no, I’m strong. I’m independent. I can do this. And then I started having issues. And I’m like, OK, I should go and check it out.
Nithin (39:31) it.
Yeah,
8 months back I got my first pair of glasses.
Federico Ramallo (39:38) Yeah, yeah, yeah. So I see the value of having a queuing system integrate with the database. I mean, as you said, you have the drawbacks of once server is down, then it affects more. But the benefit of…
of the integration at weights that dependency. then, you know, with high dependency, you can solve all those issues, right? Yeah. ⁓
Nithin (40:02) Yeah. Yeah. And yeah,
was only last week I was at a Kafka, you know, a current summit. It’s the largest possible, possibly the largest congregation of engineers and folks who work on event streaming platforms. So I had delivered a session on, you know, the value of unified. It resonated really well with customers.
And they all came and spoke to me after we had a booth as well for Oracle. So a lot of customers came and spoke to me about this exact same problem. Yeah. We are in fact having a hard time managing all this. We didn’t know that, you know, this is a product that could really help us. Let’s talk more. So yeah, I thought my end of year is going to be a little relaxed, but yeah, it has just gotten busier.
Federico Ramallo (40:48) Yeah, I mean, it’s a money problem, right? It’s good problem to have, right? Yeah. Yeah, but I can understand that that’s going to affect your holidays, Yeah, a little bit, yeah.
Nithin (40:51) Yeah, time complete.
But I’m happy in a good way so yeah as long as customers are happy I’m happy
Federico Ramallo (41:06) Yeah, yeah, yeah.
Right, right. Plus, you know, I think that once customer has this, find the need and find that it’s a good fit, then everything else is easier to do because then you can, you know, make arrangements that is convenient for both parties and you should be fine. Yeah.
Nithin (41:25) Absolutely,
yeah. Couldn’t agree more.
Federico Ramallo (41:28) So what new technologies excites you the most right now?
Nithin (41:32) So for me and my, I really like the product that I’m working on because I have seen its relevance through years. So event streaming, has, has messaging systems have been there for a long time, queuing and messaging systems. And, and this was one of my, you know, topic of discussion when I presented at that summit as well. How
things are evolving and how relevant it has stayed. Even when Kafka started, it just started as a log based system. And today it’s ecosystem supports a lot and it has managed to stay relevant for 15 years where life cycle of technology is like few months now. And today I’m really excited about the capabilities that agent-kai workloads are promising. So,
Things, I mean, you know, the pace with which you can innovate and stand up a company or, you know, start your own has been significantly, significantly affordable, I would say, with agents, right? It’s more like one person unicorn is, I don’t think it is too far away. I know a lot of my friends, I’m lazy.
to start my own. But I’ve seen a lot of my friends starting their own and really valuable products because of these agents and its capabilities. And you know, when I’m looking at my product that I handle and how it can help agent decay AIs, it’s, know, I’m seeing features that I can build, which, you know, easily fits into like pan into a glove for these agent decay workloads.
So that really excites me. So that stays that this is where the world is headed. I have opportunities to build my product, to cater to these customers, to build this agent again, and effectively catering to myself because these things are going to improve the quality of my life as well. So that is the bright side of AI that I see and I strongly believe we are headed that way. So for example, the transactional capabilities that I just spoke about, A simple example I’m telling you.
So when agent to agent communication is happening, you need autonomy. So you cannot have people just monitoring whether these things are acting properly. So for that autonomy, you need that the quality of the data that is going is consistent. that is my, that is, those are the type of things that my feature offers natively. You don’t have to think twice about it. So the beauty of building things into the database is that it, you know, offers you all the benefits of that database. So.
I reason I’m recently working with a customer, so hopefully that’ll be public very soon. So they’ve been using my product to try it out. And they were like amazed. Okay. It’s easy to use and it supports exactly what our agents require because we don’t have, we are a small company. We don’t have too many people to add a big operational team to monitor how these agents are functioning. So with these transactional capabilities, we are being able to get guarantees that okay, things are either happening properly or not.
is that not as also as important as having a yes and not to be in a gray area. So such capabilities, when I see that they are powering the new age companies and the workloads that are coming, it’s grateful that I’m in the right place. And I see that I can really build value to these, not just me, it’s our engineering team.
can build value to the next generation of enterprises and technologies that are coming. And so that’s what really excites me when I see the technologies. Yeah. AI is here to stay. More autonomy is required for AIs and these event streaming platforms are really the backbone of building AI at scale. So I feel I’m super excited about the space that I’m in.
Federico Ramallo (45:22) Interesting, interesting. when the first time you talk about transactions, was thinking on…
of the commit, right, of the transaction, if you like. But there’s also, and kind of what you were talking about now, realized me that you also can have the power to do rollbacks in the future. So if something was, know, a transaction was completed today, then you can go back in the future and say, please, you know, undo this, this particular transaction, right? While keeping database integrity.
Nithin (45:32) yeah.
Exactly.
Absolutely, yeah. Rollback is as important
as commits.
Federico Ramallo (45:53) can you share some leadership lessons that you’ve learned managing product teams?
Nithin (45:57) I’m not really good at this. Let me give it a try.
Yeah, one thing you mentioned, think it resonated really well with me. Yeah. I always try not to be the loudest voice in the room. at least a poor aftertaste for me. So I lead with data, lead with empathy. And it’s not just about system. And I really don’t like like it when, you know, your working system becomes very transactional. You know, it’s just that you give me, you give me that understanding people.
and working closely with them and collaborating, right? Don’t just make it about, you know, you do your thing and that’s, that’s the end of the relationship. No, I think even with customers, I find that, you know, having that trust based and understanding their problems and working, because if you’re just fixing a transaction, yeah, it’s short term gains for me. But when you have that relationship, it’s a sustained
growth for both for your customer and for your product. That’s what I have seen. So leading with empathy, listening to the customers, it may not always be pleasant conversations, but understanding their point, hearing them out, setting those right expectations and not just saying yes for the sake of ending that not so pleasant conversation has added more value. So I value
long-term relationships over short-term transactional ones. So that’s one thing that I have seen even not just from my days at AWS or at Oracle, even from the time I was, you know, in on the consulting side with Accenture. That’s, that’s something that I have seen as a very valuable thing. Always important to us why and not just listen to the, they want. And, just trying to build that because when you understand the wise, you always in
increase or have a better solution space to build that because they may say what they want but it may not always be true that that’s the only thing that they want or that is the right thing that they want it’s maybe that’s the only thing that they are familiar with and also accepting that I don’t know everything so ask for help when needed and shouldn’t let your or my I don’t let my ego stop me from asking for help
And in general, we create a conducive space for everyone to work together and express ideas and not push ideas down the team just because a person may be a manager or the leader of the org. Always make it two way street. And more often than not,
I learned things from folks who have newly joined and being receptive to that has really really helped me learn and grow as well. So yeah, these are the few things that I have off the top of my head which has helped me lead teams and also be a valuable partner to my customers.
Federico Ramallo (48:49) Nithin, I wanted to thank you very much for joining us today. I think we learned a lot about the new technologies. We talked about Kafka.
messaging and queuing platforms. So I’m looking forward to see what you’re taking the products next. So thank you, Nathan, for joining us today.
Nithin (49:08) I appreciate you having me here and I hope this was helpful. I had a lot of fun chatting with you as well. So I look forward to seeing more episodes of your podcast. And I want to thank you for reaching out to me and giving me a platform to tell my story. So yeah, it was great. Yeah, thank you so much.
Federico Ramallo (49:30) Thank you.