#516 Efficiency as an Unfair Advantage: Tomas Navickas on Scaling Digital Banking

[00:00:00] 

Mehmet: Hello and welcome back to an opposite of the CT O Show at Mehmet today. I'm very pleased joining me from Italy, myTU co-founder Tomas Navickas. Uh, Tomas, thank you very much for being here with me today. I really, you know, appreciate you giving me. [00:01:00] Time, uh, especially the weekend is coming and people want to chill out.

But, uh, you know, again, uh, it's great to have you here with me. As I was telling you before I start, uh, recording with you is I usually keep it to my guests to give us a little bit more about, you know. You know, their journey, you know, how you come up to here, and then we can start discussion from there.

We're gonna talk a lot about, you know, important topics today. Uh, of course about the story of, uh, my, uh, myTU, and, uh, you know, like, uh, we're gonna talk about AI fraud, cloud and APIs and everything in between. So the floor is yours, Tomas. 

Tomas: Thank you. Secondly, Matt, thank you for having me. Uh. Uh, it's a, it's a beautiful day, not just a Friday.

It's just nice sunny day and everything. And, uh, I'm very excited to be talking here, uh, speaking about myself, how I come to be at this situation where I'm today. I actually [00:02:00] started very, very early. I was into the computers and coding and everything when I was like a five or 6-year-old. Given the fact that I'm already 45, that makes a lot of, uh.

Time, right? So I was building for so long so that I can't even remember how, how this even happened, everything. But, uh, during this journey I was in various situations, you know, like a startups building webpage. I was building first the email services at similar timeframes when Google arrived and everything.

So it's. It all started at tension stages and continued to continue with every iteration. It was bigger and more complicated project, and also because I had like a interest not only in development or engineering, but also in the process as it is. And then, let's see, achieving goals. Uh, this allowed to me, for me to.

Get, let's say, training and their required education and experience to become someone like a, like a leader, like a CTO in the situation. And today it's even a co-founder. Since my last endeavor, [00:03:00] five, six years ago, I started, uh, with my partner, uh, a very complicated business. It's, uh, it's a financial institution, which, which is just another huge beast.

Yeah, 

Mehmet: yeah. We're gonna talk a little bit about this beast, Tomas, because, you know, uh, you look like you're doing, uh, great things there. So, of course, you know, I do my research and I try as much, as much as I can, you know, to, to grab information. So my team is, is known for running kind of an entire bank on less than 1000 Euro.

Per month in cloud cost, right? Like, this is one of the things that I found out. Oh yeah, yeah. That is, 

Tomas: uh, that is true. How, 

Mehmet: how did you architect this system to achieve that kind of efficiency? Because, you know what, uh, I, I see a lot of posts recently and I think there is. Complete practice now. It's [00:04:00] called finops.

So for people to optimize their cloud spendings. So how, how did you manage this? Yeah. How did you architect that? 

Tomas: Yeah, excellent question. This is just a super question. Uh, so it, it boils down to many things. Uh, it's part of my history and, and legacy. Where I came from. I was building software on the.

Kilobytes of data. So I was actually experienced in building things, uh, from scratch and building them in a very efficient way so that things use as little resources as possible and as needed. But, uh, that's not only the only thing, it's just my experience, which allowed me to design something like that, which we talk in a, in a few words now.

Uh, the other thing is that when we actually were founding the company. There was already big names, if I may mention like a revolut, you know, N 26 and others. And they already invested a lot in the marketing. It's really impossible to, to like, uh, compete on that level if you don't have like, you know, a billion, let's say, and behind [00:05:00] you.

And we didn't have, uh, so we kind of had to invent something that's, uh, that's more like a unfair advantage. And, uh, here we found that efficiency is something that you cannot copy. And if you cannot copy, this is essentially the unfair advantage that you can have because you can copy designs, so you can copy, you know, app styles, functions, and features.

And nowadays it's even easier than ever. And um, and because of that, we decided that, okay, so let's. Bank on the efficiency literally, and do everything efficiently. The processes, the people, and the software. So how I build it, the design is very specific. We chose the cloud. It's Google Cloud actually with fully Google Cloud native.

And we use very basic infrastructure. I mean, we don't build on things that are deprecated or changing a lot and that kind of thing because you would have to maintain and do changes to too many changes. It would be, uh, [00:06:00] a complication just to have something like that. So we build on the as basic things as possible.

And, uh, it's true that it was under thousand. It's, it's not anymore. Actually. We are probably having more price, uh, more cost. It's more about 2000 now if you, if think about it. But it's because we also use a lot of AI in, uh, many of our processes, uh, from software building to the actual use in the software inside of it.

So we built very specifically on a very basic tool with a very, very specific design just because we actually. Thought that we have to do it. There is no other way for us to exist in the market. And it was actually a good choice because after that COVID came, then the war and other things. Mm-hmm. And you know, it's not easy to stay afloat and having a small burn, not only on the infrastructure side but also on people helped to survive those times.

And then we are arriving now here where we are actually having revenue and we are seeing, like eyeing already break even situations and getting profits and [00:07:00] everything. This is fun part now. 

Mehmet: Uh, it indeed fun and it's fantastic. Congratulations on this. Uh, you know, it's an achievement in my opinion, you know, ha having such, uh, cost efficient in, uh, for a.

Complex, uh, use case. Thank you. It's, it's a great achievement. Now, you mentioned about AI and a couple of things and, and also, um, I'm, I'm, I'm sure you have done some optimization when it comes to, uh, support as well. So because, uh, I believe you're supporting close to maybe more of, uh, 500,000 clients. So how also were you able to, to streamline.

And optimize the, the support part, uh, from technology perspective, of course, um, that allowed you to scale without, you know, hiring more people and, uh, having all these, uh, multi-layers of support that usually we see, uh, within technology companies. [00:08:00] 

Tomas: And I asking a very good, good questions. Uh, thank you nail.

Uh, the thing is, would you believe that, uh, our support currently is the only thing that does not have really AI inside of it? Wow. It's probably the last element that we changed, uh, that we're gonna change. But, uh, at the moment it does not have. And, uh, but I can still answer your question. How can we support that many users, uh, with, with such little resources?

The answer is simple, uh, to be efficient. We also understood that our maintenance has to be as minimal as possible. That means that we are trying to build the software with as little problems as possible and not, you know, just like a automated testing like thing does not crash or something, it needs to be clear, convenient, uh, user friendly when users wouldn't have too many questions or too many things to figure out or, or something like that.

We actually employ things that even matter. Games employed, you know, like, uh, people have to. Go through some steps to explore things, to find new things, and then they [00:09:00] learn about it naturally and use that. So building the product in a very specific way because. Five, six years ago, AI wasn't on this level.

We wouldn't be able to use AI to do support. That's why we built in different kind of application development style if you, if you wish, and this allowed us to have small support for quite complicated and, uh, large user base. Nowadays, actually, you're absolutely right. Support is like a first thing that you can augment with with AI because you know, AI can have tools, access to data, it can verify, it can be 24 7 available and we'll definitely do this I think most likely next year beginning.

We'll focus on that as well, optimize that part, but not, it's not like we have a huge, uh, load at the moment because of the strategy, how we build it. 

Mehmet: Great. Now. You are in a. Highly regulated, uh, you know, [00:10:00] industry, right? So, and one of the things, uh, especially when anything related to to, uh, to finance is fraud, right?

And, and we hear horrible stories about fraud. I've seen like you, you've written something related to, to fraud prevention and the role of AI in, in, in the whole thing. So let me start from, from AI because it's, you know, top of mind of everyone. So how AI is changing, you know, the way. I would say banks detects, respond to fraud compared like maybe to, uh, previous traditional methods.

Tomas: Mm-hmm. Yeah. Again, that's very similar story to support, but, uh, here we started using AI and automation way, way much earlier. Uh, the thing is that, uh, fraud prevention does not happen with, let's say. Just detecting or having some special tools. It's [00:11:00] also about designing the experience. For example, uh, if users, uh, don't have logins and passwords that they can forget, that's, uh, a thing that will or, or lose.

Or if there is no web access with a login or password or something like that, which can be phished from them, they could leak something like that. So, uh, account cover fraud types are almost nonexisting with our solutions. It's because it's. Practically impossible or too difficult to execute, and that's why simply nobody takes time to do it, because fraudsters also are businessmen considering that they spend time where they can earn on it.

If it's too much hazel, they, if it is too complicated, they simply don't do it. That's one thing. Another thing, yeah. So once we started doing, uh, let's say, uh, more transactions, uh, the fraud comes from different sources. Not just like, let's say, says parties, but also from within insight from our customers also.

Customers might defrauded by, let's say fake merchants and [00:12:00] similar kind of things. So we built in AI detection, like literally first rules. Then ai. AI is more flexible because we don't have to be too much specific about conditions that need to stop this operation. For example, a customer receives a payment.

Let's say our customer receives a payment and suddenly he wants to make that payment. Transfer the money somewhere else, maybe to his own accounts. And, uh, there could be like, you know, red flags, let's say false name usage or, or vague reference messages or something like that. We could stop this transaction because we think that, okay, this our customer could potentially be fraud.

That he's frauding someone with a fake, uh, sales that he's not actually doing, and then sending money away and, uh, making, uh, specific rules for this works to extend, but they. Also, uh, well organized. They share this information between each other. They have groups, close groups, forums where they share what kind of provider has what kind of, let's say, thresholds for what, and they can pass that [00:13:00] information, split transactions and everything.

Having AI is very different story because AI does not have strict rules. You do not make like a program do this if it's that you make. Like, uh, guidance. And then the LLM decides based on the guidance and the knowledge that it has and maybe some other knowledge that we add. And with the tools, if it needs to read some other additional information, it may make a decision to pass or stop this transaction.

And this gets way harder to, let's say, fake Another thing, let's say transaction is stopped and we ask questions. AI can prepare questions in advance. That's one thing. Another thing is if the client responds. It can validate authenticity of the data it provided in a way that humans cannot. So it really changes how fraud is handled.

And, uh, it's where we actually have something because we catch a lot. Yeah. And to enter the fraud series, catch the, when you catch him, he's not too much happy. He goes and, you know, [00:14:00] writes bad things about you on, on the many platforms as you can and so on. So that's another thing. Yeah. 

Mehmet: Yeah, it's another thing, uh, just, you know, a small, a small, uh, thing about what you mentioned.

I had a colleague, uh, who I'm still friend with him, so Exactly. He used to tell, uh, we were selling something related to cybersecurity, right? So, mm-hmm. He was telling them like, uh, the bad actors are people like us. They have bills to pay. They send their kids to schools, and they have to be like, you know, they have to be like also creative in, in, in finding new ways.

So, absolutely. You know, this is why also, uh, we see them, they are always leveraging the new technologies also as well. Mm-hmm. Uh. We, you, we talked about AI and you know, what it, how to implement it and how, how you did it yourself. But do you see to, you know, to be playing role more in, uh, proactive [00:15:00] prevention or predictive one, which is mean like proactive.

You know, like, you know, before or after. Very interesting. 

Tomas: Yeah, that's very interesting question. I think that, uh, since AI and, uh, LLMs in general, they, they are not sentient beings and they do not have constant input like we do. We have constant stream of information going. Interest, right. And that's why we operate constantly unless we're sleeping and even then we still operate at some level.

Uh, so AI does not do this yet, and uh, you have to feed it information. So I would say it's reactive, not proactive, but it reacts, uh, on the events that are happening like now and given the timeframes that we are allowed to, to, to, let's say spend on some things, uh, we can actually do quite a lot and make a decisions in real time.

So that user experience is not too bad in terms of how quickly payment goes through. 

Mehmet: I want to go back a little bit to the cloud story, [00:16:00] but this time not, uh, about the cost efficiency, uh, rather about, you know, uh, maybe to inspire, uh, uh, banks to do something. So we know like traditional banks, um, they struggled and maybe they're still struggling with cloud migration.

So what's the edge you've seen? Building, you know, digital bank cloud first versus trying to retrofit, you know, a legacy system probably running on, because at some stage I was, uh, you know, I did manage banking systems, but I was talking to people in banking and, you know, like these, uh, uh, monsters and beasts that we call them mainframes and all, all the complex system on top of the head.

Tomas: Yeah, they still use that actually. Well, another excellent question. I see you are prepared. So the thing [00:17:00] is, uh, that, uh, for banks, it's uh, it's a no win game actually, because they have such a big systems, you, you, you cannot just. Take it as it is and transfer it to the cloud. It'll not work for many different reasons.

Starting from security to the governments, to the scales, to the data amount, to everything. It's, it's a very, very complicated thing. And there will be no even benefits of having this in the cloud, uh, apart that, okay, you are free from the hardware that you used to have, but you still have to maintain everything, you know.

Have a lot of administrators, network security, DevOps, NetOps, a lot of things, and, uh, it's a very difficult game for them. They try to mimic this by creating, uh, let's say user interactions edge, for example, of the mobile apps, web apps in a different way. Uh, and they use at the backend, uh, let's say replicated data, not even directly from their systems as it is, but [00:18:00] replicated data from the, some middleware, which has the replica of everything from the core and so on.

It's so complicated. Nobody really knows everything inside of the bank, how it works, and therefore it's really impossible to transfer it or. Even rewrite it. The only way to do it is actually a clean state. Some banks actually decided to do this. They did a spinoff, let's say, or either buy a FinTech or create of their own and do everything digitally.

Ground up, new team, new, new people, everything just separated again because they have a legacy of years and years, decades, possibly even more. No one knows everything that's happening inside. 

Mehmet: Just a question that just came to my mind, uh, Tomas, like, uh, related to, to, uh, you know, the architecture of, of a, uh, digital bank, do you think, um, like if the cloud, you know, [00:19:00] wasn't there.

Would, how hard do you think it would be to come up with the digital banking model and, and you know, of course, uh, the model banking. 

Tomas: Okay, this is, uh, still possible. So the thing that cloud gives me, which I wouldn't have if the cloud were not there. So cloud gives me ability to use some, let's say, managed services, right?

So I don't have to have network specialists, I don't have to have workstation service specialists who are patching everything and so on. I get the basic. Platform as a service, let's say. And uh, if I were to build everything of my own, I would have a higher capital investment needed. I have to buy everything I need to buy, not one, but multiple sets.

I need to have several sites because of the, you know, resiliency and everything. And I would have to have personnel here and there and further. The efficiency levels that we have are simply not possible, and uh, it would mean that the [00:20:00] cost upfront is way higher. I wouldn't be able to have the startup I have now.

Mehmet: Yeah, a hundred percent. And of course the speed of, uh, you know, bringing things, uh, from, from maybe test and dev to production as well, you know, having this flexibility of completely. Replicating and this is why, you know, this brings uh, me to the, the next question. 'cause you mentioned about security measures.

We talked a little bit more about the fraud. And actually you have to do it whether you are a normal bank or like a digital bank. But when it comes to cloud still, you know, there are some question marks, especially in the banking sector about. Sometimes data resiliency. Like I know like you're based in the eu, so probably Google Cloud is in, uh, Europe, uh, zone, I mean in region.

So you don't have this problem and all your customers are from [00:21:00] Europe, but data, you know, uh, sovereignty also sometimes, and security, right? So, uh, is it a myth? Do you think that the banks still are believing that the cloud. Isn't secure as the data center and you know, the measures that we do for on, for our on-premise workload, we cannot get it in the cloud.

Of course, I'm sure they know about all the optimization you mentioned and the speed. You can bring things up, but majority of time when I speak. To people who are still kind of in traditional banking from technology perspective, like these are the first two things that comes. Of course, they mentioned other things also as well, but mainly related to data security, privacy, which related to how cloud operates, right?

So, yep. Is this a myth or no? Like actually. Not true. [00:22:00] 

Tomas: Oh man, this is a huge question, uh, to answer what I think, what I believe. Sure. I think this is a myth. If it was, if I didn't think that it is a myth, I would probably not use the cloud services here. So in general, it is a myth, but, uh, I don't believe that, uh, uh, others, uh, let's say CTOs or even let's say groups or in departments and the banks actually believe that it is insecure.

Or something like that. I think they understand the limitation, what they can or cannot do, and they try to move as many things as they can for convenient, for convenience to the clouds. But they sometimes have, let's say, extended clouds, like uh, the same cloud infrastructure with a managed service, but on their premises and their hardware.

Mm-hmm. It's possible to buy that kind of services and, uh. What is true is actually that policies that banks have is, uh, very restrictive on what it can actually do. The it's and the policy will not be changed, you know, in a day or a year, or [00:23:00] possibly even in a decade. That needs to be some, some huge reason for this.

That's why they have. Everything they need in-house, they need to have complete control on the data they, on their business, on their data. It's billions upon billions we're talking about, and they don't trust anyone else to have it. Since they were not built originally in the, let's say, public service, where you have, let's say, solved this, uh, data control thing when you have multiple copies, uh, real time log shipping, and you have replicated data like on application level if needed in real time.

You control it on a different entity, and that's how you manage your risk and dependency on your service provider. Because you know, if you are a cloud, if you bank in the cloud and the cloud suddenly decides that, okay, we need to throw you out, there is nothing else. There is literally nothing you can do to stop them.

You have to have your data somewhere. And, uh, they don't believe that they can [00:24:00] solve it in many cases, or the policies do not allow them. That's why they will not also migrate to cloud completely. So it's not trust. I think policies is the problem here. Uh, better for us. 

Mehmet: Yeah. It's, it's opportunity for you.

Definitely. Like, you know, um. I, I, I always tell people like the biggest opportunity, regardless of, uh, the vertical is if, if someone can do exactly the scenario you mentioned, like if the, of course this will not happen, but let's say one of the cloud providers decided like, okay, sorry guys. I'm, I'm not, uh, supporting your region anymore.

I have to close the shut down the data center. So I, I always tell people like. The one who comes up with a way that you can migrate seamlessly workloads across different clouds. That would be a great things to do, really, because, uh, uh, especially for, for, for the people who are still, maybe [00:25:00] even on-prem, like they tell me, usually they used to tell me, uh, is that if they find a fast way to bring workloads from the cloud to on-premise, maybe they would do it.

Uh, but they feel that cloud providers, you know, so, so, you know, they blame and, and, and sometimes I don't, I don't disagree much with them, but. Also like cloud providers and which is their right. Of course, if I sit on their side, they try to do what's called vendor log in, right? So you are, you are relying completely on that ecosystem and it's very hard for you to leave it.

And some people who are like kind of, they have the, they want to have the control in their hands. They say, Hey, listen, like I, I, I don't want to be like a hundred percent tied to you, just like food for thought, right? 

Tomas: Yeah, it's all true that you see, uh, the thing is that, uh, if your system uses a lot of different cloud services from [00:26:00] single provider, you are inherently dependent on, on, on that cloud provider because moving anything will be hard.

The more different services you take, the harder it is. That's why we build it as, uh, simple as possible to be. Agile if needed to move from one cloud to another cloud on premises if needed. That being said, cloud providers also understand this. That's why they have, uh, let's say, products, uh, tailored for that kind of use cases.

For example, it's not a Google advertisement, but I, I've been there, talked to them. They have a product you can buy hardware, which literally has a parity level and functionality of cloud service. You get on premises, Google Cloud, but on premises. 

Mehmet: Mm. Yeah. 

Tomas: And everything just works the same way. So all the tooling and everything, and you, and you have the hardware and the software running in your environment.

So that's possible. It's very expensive, but that's possible. Another interesting thing that I saw actually, we, we haven't even touched AI a lot in, [00:27:00] in this conversation yet, but, uh, I know that, uh, bigger banks, uh. Have policies that prevent them from using no public ai and for, and probably for good reason because, uh, it's a question of education.

If your employees will post somewhere where let's say some letter, some information or something will be logged and then used in training date or something like that, it can happen. That's why they block everything, because they have a lot of people working there and not everybody's well educated on what.

You can do or cannot do, and they provide their own internal services, including LLMs and, uh, LL. Those LMS are allowed to access, let's say, private data in the even customer data and similar, uh. Similar, those are similar services to what you could get from public. It's not like you cannot have safe and secure public services.

It's also possible it's, uh, mm-hmm. It's not really a complication. Uh, but they decided to build everything on of their own in, [00:28:00] in-house because that's what they understand and control. Get question of control and trust. 

Mehmet: Yeah, it is. It is the control and the trust. And you know, my point, and I take it maybe from cybersecurity perspective more because I tell them like, guys, like if, if someone can.

Get into your systems in the cloud. So, because it's not 'cause of the cloud, 'cause of you. I'm talking about in general. Yeah, most likely. 

Tomas: Yeah. All true. Yeah, very true. 

Mehmet: Yeah. Yeah. So, so you had, you had problem with your identity security. I don't know, like maybe you have, uh, uh, some misconfigurations, which can happen on premise, so it's not the cloud of course.

Um, so, so a hundred percent on, on this point. Now, one of the things, uh, also you, you, you're driving innovation in as. Per my research is, you know, um, B2B payment through, uh, APIs. Right? Yeah. Uh, and APIs, you know, are [00:29:00] reshaping, especially in the, in the, you know, in the FinTech are reshaping the way, uh, businesses handle payments.

So tell me more about it, like, um, how you're leveraging this, what kind of innovation you're doing. 

Tomas: Okay, so we actually do have sort of innovation in API levels, you know, uh, actually. A year ago, we didn't have even a product for businesses over APIs because I kind of thought that this is well served already on the market and, uh, there is no need for something.

Well, because, well, you know, you can get it from everyone, but as it turns out I was wrong. Uh, it appears that, uh, a lot of service providers, uh, specifically banks and even digital banks, like modern banks have problems with APIs, and some of them are from the performance level. For example, if it's API served by the.

Big bank, they may have problems with the queues. They stop to process payments or certain days or [00:30:00] hours. Uh, and, uh, for businesses that need 24 7, uh, payments going in and out, this is actually a serious problem. Uh, another thing is that APIs need to, uh, like comply and provide proper security and comply with, you know, strong customer authentication as well, not just, let's say.

Basic authentication that many API implementations actually have, you know, you have like a token and you can just make a payment and that's it. It's okay. It's, it's good if, if they at least, or at least CP then it's another layer of protection. But, uh, what we built, we built a very secure API with the having separate two separate at DEA is, uh, validating DNS records, validating IP addresses coming in and out.

And we have call, let's say, uh. Multiple steps in, involved in the API and you would think, okay, so this API is seriously complicated. Nobody's going to implement it quickly or even implement it [00:31:00] easily, or something like that. And to some extent it's true because the quality of the developers, it's uh. Not always on the level that they are con like, let's say, comfortable with their custom things, which are not like, say what they saw before.

But, uh, we also found the, uh, the opposite. For example, we have a client who implemented everything and transferred all their payments to our platform within a single day. I mean, we set up environment for them. They do development and start running next day. One day to do everything. So they had a problem.

The operator failed to provide their services. They were offline for a few days, and, uh, since we already had a contact with 'em, they contacted us. We said, okay, so let's do it. We can support you, let's say developing quickly. And they did everything in a single day. So it's not like a complicated API means that it's, uh, it's a block.

No, it's a question of, uh, well, if you want to spend a little bit of time to understand it and implement it, it's absolutely [00:32:00] doable in one day. And once it's done, it's super secure, super reliable. And then it, it was built naturally in a way that can actually support things like verification of PE automatically.

Like, uh, we didn't have to change anything important in our API to have this new thing that that's actually mandatory for everyone. And now look at the businesses. For someone who, who's doing batch payments or APIs, they have no feedback. They just have a post and the response and. Is it, so they have to find the ways to explain the regulation in a different way.

Explain everything to auditors. No, but our APIs are very simple and they build in a way, in, in a way that where verification of PE fits naturally. 

Mehmet: Talking about, you know, the API still, but I want to, to, to blend it with the ai because you mentioned like sometimes people will not have the, you know, the, the skillset, uh, needed to, to build on top of that, you know, [00:33:00] do you imagine with the API and ai, LLMs and all this, that.

We can, you know, change the whole way of how these, uh, BB business payments, like B2B payments are handled because, you know, first I can get an AI to build for me the module on top of the API second, maybe even better, I'm not sure if I'm talking, uh, something very futuristic. But you know, convert this into kind of a chatbot.

So for example, like, hey, like this is the API, you know, you have to use, of course you have to give still the tokens and everything. Of course the security measures. But like building something like, Hey, go make. Payments to this and it'll go find what's the API function that it has to use and do it on the fly.

Am I talking too much futuristic or, 

Tomas: uh, it is not futuristic. It's actually what is possible today. I wouldn't, uh, say that AI would have to understand [00:34:00] the API and build on top of it, particularly for payments, because it's a multi-way communication. You have to expose some end points to receive callbacks, to, you have to make signatures.

It's question of how, what. How and what kind of tools you give to the LLM and, uh, it's absolutely possible to give a payment tool and the tool itself will have integration to the environment it needs to actually execute the payment. First of all, it needs to make a PA call. Then it needs to make responses on, let's say validation on signatures and so on.

And then it needs an endpoint to accept, let's say, progress, right? So how the, the payment is going because it's not a, a real time in terms of like HTTP request. You make request and you get response. It's as synchronous because you post your order, then it's being processed and you get callback, uh, progress reports on this.

So if you give a. Tool, which is basically make a payment from certainly banned to any event, blah, blah, blah. Confirm it. And it definitely will do it easily. And you can, let's say add this, [00:35:00] uh, ai, make it, uh, let's say automatic invoice payment ai. You add the channel, for example, you receive invoices from someone.

AI can make payments automatically on let's say same day last day of invoice expiration, wherever you need. And you can have pre-approved set of, let's say, where you pay automatically because it could be a way of hacking you. And that's this way actually will, problems will arise. Somebody will build AI to do exactly this and will not, let's say white list where the payments can go.

Somebody will find that there is a point email where you can just post your invoice and the money is coming. And, uh, these things will happen because if everything is built automatically, if the there is no, let's say, good overview on everything from the security logical perspective and everything, it, it can happen that there are mistakes.

And, uh, with LMS nowadays, that's actually the tendency because what I found myself working here because my personal work and life changed a [00:36:00] lot. In the past years, like, uh, wow. But, uh, what I found that the LMS are quite lazy in doing things. They skip things. They pretend to do things by placing placeholders and so on and not implementing everything.

And, uh, this is a serious issue at the moment because AI builds. Uh, AI is very potent thing, but it needs very good supervision and guidance during this. And this is where good developers come in because good developer can use this tool and build excellent software like 10 times faster than he did before.

But if you are just a junior who is guiding AI to build something, he will be happy with what you get to see. And AI is very efficient in making, uh, cheap and optimal solutions. Which skip the important things. Absolutely. Just have this as your idea, like it's, it's today's reality. I hope it'll change. I hope it'll be more [00:37:00] thoughtful.

About what it is doing, but at the moment it is probably because of the reinforced learning techniques that are used in training. It's very goal oriented, and you can achieve goal by, let's say this is a problem, I have an exception. It can remove the exception line. There is no problem, right? Exception is not thrown, but it's not a solution actually.

You just solve the symptom, and that's the tendency of today's lms. 

Mehmet: Um, you know, I was talking to another CTO and he, he was mentioning something similar to this and, you know, he was a little bit, uh, not, uh, happy. He wasn't happy with, you know, because that the LLM allowed what's called now vibe coding, right?

Mm-hmm. And you know, so the amount, yeah. I tried multiple vibe coding tools, like all, all, almost all of them, right? Without mentioning names. So no one get pissed off of what I'm saying. They're good. Don't get, don't get me wrong guys. Uh, and I'm not a [00:38:00] developer by, by any mean. Okay. I had like, uh, 20 years ago some courses in Python and c plus plus, but I mean, I'm not a developer by any means.

But what I noticed, to your point, and I think this is not only applies Tomas, for coding only, even like when, when you give the AI a task, it tries to, it tries to be lazy. Especially, I noticed this after. The LLM has been released out for a while now. One of my guests told me they do this on purpose because they want you to, to upgrade to the next level of the your subscription.

Someone told me you never know me. Yeah, you never know. It's like only possibilities. My experience the only. I'm not, uh, affiliated with, uh, with any company of AI companies. The only one that I didn't see degradation, although like they used to take time to release new ones, was cloud the Tropic, um, LLM model.

But all the rest from Gemini [00:39:00] to Chad, GPT, uh, I'm not mentioning the other ones. I mean, at least these are the three main ones and. Maybe now grok from, from, from X. But you know, Claude is the most consistent one. Uh, and when it, it doesn't want to work, it tells you, Hey, like you filled your, uh, your limit for today or for now, come after, come after four hours.

Chad, GPT on the other hand, keeps playing this with you. Yeah. Okay. Okay. Like, I've seen this, let me do that. Let me do this. But, but, but you're right now regard, you know, other than co like, uh, helping in coding and helping in, in, in building, what other, and you mentioned at the beginning, of course about the optimization, what, what you've done in the, in the cloud.

What other use cases are you seeing which are fit for fintechs specifically from, from AI perspective?

Tomas: I think, [00:40:00] uh, from FinTech specifically, AI eventually can replace everyone. It's a far thought, like a long shot, but, uh, eventually. No one is needed in a FinTech, not even CEO. You can have zero human involved in something like financial service here. It's fully possible to automate everything from support from MIL compliance.

Compliance is absolutely way, way better than humans because it can dig into things deeply. It can create risk reports very quickly, very precise, very like, uh, it can take into account things that. People simply cannot fit in, in, in minds. And uh, I think for FinTech, if you consider what, what FinTech is actually doing, you can replace absolutely everyone.

And it can be fully digital, fully automated service. It's not like I envision that we will become something like that in the near future, but it's quite a possibility. The only thing that the regulation would not probably understand [00:41:00] something like this because, uh, who is controlling who? So this agent is controlling that agent and that agent is controlling them and this is that?

No, it's the chain. It's even from, from, as I understand that. AI and LMS are suited perfectly to execute exactly what, uh, let's say supervised entity has to do because you can supervise everything on every step. One of the things that we do, AI is actually monitoring communication with our clients and it can prevent, you know, like, uh, things that are offensive to any direction.

I'm not even speaking about the translations that it does on the way. Catching context, which, uh, you know, people write with mistakes in slangs and so on. No translation was possible before the LLMs. With the modern lms, you can understand and catch the context, what was the problem and frustration of the client and explain this to the operator.

And you can catch if operator is doing something incorrect in like a, what he shouldn't be doing, contacting [00:42:00] the, the client. So, um. Like I said, everything in FinTech can be the first things that will be changed is of course support, MOL and compliance, I think development enhanced. Mm-hmm. Not replaced, but enhanced now.

Next, probably replaced and uh, I will have no job. That's great. 

Mehmet: Uh, so, you know, like should, should founders and maybe people who are considering, uh, FinTech careers be worried now? 

Tomas: Uh, no. There is no need to worry. It is because, you know, it's like, uh, every time the industry revolution comes, uh, a lot of things change, right?

Of course, absolutely. Jobs will be taken. Absolutely. Something new at the moment is created. For example, you have to have. Basically AI supervisors or AI core developer, developer or something like that. It's different kind of position emerging. In the market that, so [00:43:00] we need someone who is actually able to talk to those things and explain tasks.

And as it turns out that the senior developers actually is a good, uh, fit for even for not for development positions because they had the history of explaining things to people, to juniors, to businesses, to clients, and so on. They're very good at in talking with LLMs and convincing them to do what they want.

That's why the, uh, perfect material for any kind of, uh, ai, prompt engineering jobs, like related, something else will come in the future. So there will always be something. Oh, just remember the horses when the machine came, everybody thought that, uh, okay. So horse industry is destroyed. To the extent, yes, but there was machine industry, like a factory built and there was new places for everyone to work and the horses didn't went to extinct.

It's just a luxury service. Now you can ride horses and everything, so things will change. People will be luxury in the digital world, let's say. [00:44:00] 

Mehmet: Yeah. 

Tomas: Uh, but, uh, there will be a need for them because not everybody will want to, to talk to support on the, on the ai, but I think, uh, even that will change in some time because generations are changing and, uh, current generation is so, so used to this that they will, they will not know that it was any other way sometime ago because they, they get services instantly on demand and, uh, it wasn't available before.

A 

Mehmet: hundred percent now. We are talking about, you know, what have been done, you know, in FinTech. So I remember when the first wave of, uh, digital banks sometimes called nail banks and, you know, came, it took the war by storm because, you know, uh, it was really a revolution because if you think about it just.

Opening a banking account was, I don't know, maybe a one week process, collecting papers, going to the branch [00:45:00] waiting, and then until they send you the, you know, the card physically and all this, and then, you know, this digital branch or digital bank and you know, neo banks, you know, it's just like matter of, uh, minutes and you are up and running.

You have your IAnd number, you have even your, uh, uh, virtual card before the actual card comes and you can start to use it. So this is this, this is a big move. So if, if you want to predict, you know, and of course predict in a sense like of not a hundred percent, but what do you think like the biggest next innovation could be, uh, in this sector?

Tomas: Okay, I need to actually think. Sure. The biggest innovation I predict, uh, that the biggest innovation will be. The user interface, the experience that a client gets, uh, interacting with the bank. So at the moment we have, we had branches. So you were going and talking to the teller, and [00:46:00] then you received some documentation signing out of papers, then you received the card.

Then when, when you want something to call, make, appointment, go, it changed a little bit. Now everything is possible online with digital identity verification methods. It'll be even easier to validate your identity just in a couple of years when the, you know, a, those regulations come into force and everything.

Uh. So I predict that, uh, front end of the digital bank will be single box. There will be no buttons, no list. That will be chat window, just like, uh, any other interface. This is probably the near, highly advanced future. 

Mehmet: Or maybe voice, right? I can just, 

Tomas: that's next level. So once the devices, uh, so once you get accustomed to, you know, conversational, let's say interfaces, right?

When you are talking to your application through the chat or through the voice recording messages, it's basically the same because now [00:47:00] voice transcribing is like optimal and it can even be done in real time. So what you speak is literally what you type and, uh, right, whichever is faster for you. I saw how young people type, they type like a, I cannot read that fast.

Uh uh. But, uh, once they get acquainted with the idea of conversational user interfaces, they can actually transition to the screenless interfaces. Probably it's possible that in the future it, the phone will be something that even has no. You know, screen or monitor. And if you want some graphics, you will probably cast it on some projector in on around you or something else.

Or maybe in the glasses that you have and wearing it, they can project the screen to you. But without that, the phone will be just a box that you're carrying around or, or, or, or something like that. And you will be just talking to it. Eventually far, far ahead. You know, if everyone gets their way, maybe thinking about it, but we are talking about [00:48:00] sci-fi here.

Mehmet: Uh, you know, I don't see it as a sci-fi anymore, you know, from, from, 

Tomas: yeah. You would not believe like five years ago or 10 years ago that No, there will be chat interface as you will talk and you will not be able to identify if it's not human responding to you, blah, blah. Yeah, it change a lot. 

Mehmet: The first time, you know, I noticed like it was like two days after they released it, the start, people talking about Chad GPT.

I said, what? Going back to a chat interface like this is odd. Do you know? Yeah, 

Tomas: yeah. It's a full cycle. 

Mehmet: Yeah. I just said, okay, hold on, let me try. Uh, and first I, I thought it's a joke. I thought like someone was doing hype for nothing. I knew about open ai. And then I start to think, okay, what's the relation, 'cause these guys on their website were mentioning something else.

And then anyway. The first interaction, I said, oh wow, this is, this is next level. [00:49:00] Um, and you know, all, all my guests, they agree, you know, on, on one term, like the ginny is out of the box. So now really it's, it's just our, uh, imagination where, what can we do? With this technology. Of course, still there are some fact, some stuff it's far, uh, because I think it needs like more time.

Some of them you mentioned also like maybe the regulation needs to be, uh, in place. So we can bring that up. Yeah. But anyway, um, so Tomas, as we are almost, you know, come to, to the end, um, as a CTO now, and as someone who. Who, who came, like, you know, all the way to where, where you are today. What's your advice for, uh, fellow founders, co-founders, uh, in this space?

Uh, if they are maybe just now starting, or maybe they are still, you know, drawing things on the napkins, so what would be your, your word of advice to them? 

Tomas: Draw on the napkins. That's a good thing to [00:50:00] organize your thoughts. I like that. Uh, the best advice, uh, don't start another FinTech. It's a very complicated business, so don't do that.

Do whatever you like and everything, but uh, just be open to the ideas. But don't be crazy about it. But be open to the ideas that, uh, might be you don't need. So many people or specialists around you to build things you can actually build. Now with having, uh, an AI as an expert next to you, you can have, for example, if you would think that, uh, you need some help from someone, most likely you can just get this help from cloud or chat or, or Gemini.

Most likely. You can't get any expertise you need. You just have to learn how to ask questions. So. Important skill, new skill you have to have is prompt engineering. Learning how to ask correct [00:51:00] questions. Not, it's not the question, can I do this? No, you cannot. You have to ask how I do this? What do I need to do this?

What, what I need to, for me to be enabled to do this? That's kind of questions you have to ask. Uh, it's just the beginning and then you dig deep and then get your answers and, uh. You have to try a lot. For example, I achieved, uh, a desired goal only on the 17th, no Teenth 17th attempt. So 70 attempts went by, and the last one was like a, for some reason it was like inspirational.

The AI came out with the thing I needed. Wow. For, so for, I don't even know exactly why. Maybe it went, the request went to some experimental alarm underneath or something like that, but it was stubborn on do not achieving what I needed. But on the 17th attempt, it was like inspired and it created something once.

So also don't give up. Okay. On anything you do, because there is the only way to lose. You either win or you learn, [00:52:00] you lose only when you give up. 

Mehmet: Love this. You lose only when you give up. Absolutely. And I think, you know, just, uh, as a, uh, small remark on what you said, I, I think this is also because LLMs kind of.

They are programmed. Right. So they are programs and they are programmed in the same way as, as humans? Not really. Not really. Yeah. But sometimes, sometimes I've, I've, I've seen this myself, like, yeah. Uh, I ask the same question, not within the same chat by the way. I open a new, a new prompt. And I ask, you know, the same, I tweak, you know, my prompt a little bit, actually, I, you know, I vibe coded.

Uh, and this is for myself. It's not like for just a, a prompt, enhancer or optimizer. So, uh, and, you know, since I start using it, I'm, I'm getting results pretty much good. But anyway, so Tomas, um. Where people can get in touch if they want to learn, maybe get in touch, maybe of also they [00:53:00] want to about my, myTU, um, so where they can find you and read any article that you might post.

Tomas: Yeah, thank you. My, uh, so the best way to contact me is probably LinkedIn. I respond on LinkedIn messages, uh, specifically if it's, uh, I don't know, industry fellows or something like that. They want to chat. I will chat, uh, or this is it. LinkedIn is the only and the best channel to contact me. 

Mehmet: Great. So I will make sure that, you know, the, the profile link is in the show notes, so if people want to reach out to Tomas.

Tomas, I really enjoyed the conversation. Like Yeah, me too. Thank you. I eyeopener, I would say for, for tech leaders. And even people who are interested in FinTech in general, and, you know, uh, the AI fraud prevention, security cloud optimization, and all the topics that we discussed today. So really thank you for, for the time today and this, this is how I end my episode.

This is for the audience. Uh, if you just discover this [00:54:00] podcast. By luck. Thank you for passing by. If you enjoyed it, give me a favor, subscribe and share to your friends and colleagues, and if you're one of the people who keeps coming again and again, thank you very much for your support and for your, you know, efforts, uh, making the podcast trending in multiple countries at the same time.

This cannot happen without your support. And as I say, always stay tuned for a new episode very soon. Thank you. Bye-bye.