This article first appeared in Digital Edge, The Edge Malaysia Weekly on January 27, 2025 - February 2, 2025

The newly launched Ryt Bank — a joint venture between YTL Digital Capital Sdn Bhd and Sea Ltd — is entering a market that is seeing stiff competition. The bank aims not only to capture significant market share but also to out-innovate its competitors by providing the best products and services to Malaysians from all walks of life.

Taking an unconventional route, the bank is focused on using the wide capabilities of artificial intelligence (AI) to build a digital bank that is easy and intuitive to use. It aims for users to interact with their computers when using its app, just as they would with another person.

“For example, telling the app ‘Pay Mike 50 bucks for lunch’ and it will understand and execute this command. This is one of the things YTL AI Lab is doing: reimagining the human-computer interface for banking systems,” says Foong Chee Mun, chief product officer of YTL-Sea Digital Bank Project and CEO of YTL AI Labs.

“This is the first time in computing history that computers can understand human language and not force humans to learn computer languages.”

With 24 years of experience working with AI technologies in two companies he co-founded — Simulex Inc and MoneyLion Inc — Foong was handpicked to spearhead the digital bank’s product development by Yeoh Keong Hann, a third-generation leader and executive director of YTL Power International Bhd (KL:YTLPOWR) in 2023.

Foong was on a career break at that point but chose to end his hiatus earlier than expected to take on these roles within YTL. “The truth is, the break was really nice. I got to spend time with my family and I learnt it’s really about the quality of time spent with them.

“But after a few months, Hann came to me and said, ‘Hey, I need someone to help me build the [YTL-Sea] digital bank and you’re the only guy in Southeast Asia who has actually built a digital bank before’. He was really convincing.”

The pair shared the same vision — to utilise the latest technology and AI to build a bank and a mission to use Malaysian talent to develop the digital bank. Foong’s in-and-out understanding of AI and its capabilities presented an array of possibilities before Ryt AI. Working within YTL AI Labs meant that they could continuously work on improving the banking tools and interfaces, making them seamless and more intuitive to use.

The bank’s primary challenge in the coming months will be retaining its user base.  The advantage of a digital bank is that it has minimised the barriers for customers to join, says Foong. However, this ease of access also means that customers can leave just as effortlessly, making customer retention a critical focus.

“We want to be customer-centric and provide good service and good value to the customers, which is why we are investing in AI to create tools for customers,” he says.

“To say that the digital bank will help you save some transaction time might not be true but what we want to do is to provide tools that can make it true or at least lower your cognitive frictions to get banking services done.”

This is also why the team started working on is high-frequency tools, such as for payment. “We are trying to make payments and transfers to be done in natural languages, so if we help you save time, that is the best and most basic value we can give anyone.”

Moreover, the company is determined to take a minimalist approach with the number of talent used to build the digital bank, ensuring that only the best talent is recruited. “That is something we did and we’re proud to say that we managed to build the digital bank with about 25 local engineers. Most banks have more than 100 engineers,” says Foong.

Why is there such a significant gap in hired talent? Foong believes it’s because the country is at the intersection of two major technology trends: AI and low-code technologies. The advent of large language models (LLMs), which excel in language processing, has particularly enhanced programming languages. These models enable programmers to be highly efficient at any given time.

“We’re talking about 40% of the code in our repository is generated by AI, so that is a big productivity boom for us.”

The rise of low-code technology allowed them to build internal tools without going to software-as-a-service (SaaS) players, which can be pricey, especially if there is a need for system integrators to customise tools for the company’s needs.

“We built it from the ground up and managed to build the whole CRM (customer relationship management) system with one engineer and without a SaaS player. To me, these are very big technological trends that I think will grip the industry in the very near future. Not just for the banking industry but for any other industry as well.”

While there are already other digital banks, Foong isn’t too worried about the competition in the industry as he believes each bank has its own strengths and target segments. For example, Grab would probably intuitively garner its revenue from loans to its e-hailing and delivery riders as well as doling out products and services to the platform’s merchants, which Grab will have an understanding with.

Aeon, on the other hand, has been making loans for more than 20 years and it would make sense for it to focus on that area. As YTL has invested in AI and partnered with Sea Ltd — who has a significant buy now, pay later (BNPL) market — to roll out the digital bank, Foong says the focus will be on YTL’s business ecosystem.

“Just the ecosystem around YTL and SEA itself can be quite huge and will be able to turn us profitable as well. In the near to medium term, all the digital banks will mostly not compete with each other and if you look at the ecosystem, we also won’t compete with the big banks.”

When the computer first came about, humans needed to understand its computational language in order to give it commands. With generative AI, this has now shifted to the computer understanding our natural language.

Understanding this, Foong says he wants to make the banking experience more human-centric. The ability of LLM to do the maths presented an interesting use case, which formed the basis of how the Ryt AI, an AI-powered private banker, was built.

“So, let’s say you go out for lunch and you’re too lazy to calculate; you just tell the application, ‘Pay John 56 bucks divided by two’, and it will be able to understand, calculate and transact for you. Essentially, it’s like you’re talking to a human being, a banking butler, so to speak.”

While this may seem like a tool to make us lazier, it would significantly lower the barrier of entry for digital banks, especially among the older generation who are generally averse to technology. The goal, says Foong, is to develop the tools well so that they would be more likely to use it intuitively, adding that Ryt AI is also able to converse in Malay.

Foong is hoping to bring in technology that was created during his time at MoneyLion, such as supervised learning for banking back office-type transactions, for example, credit underwriting. MoneyLion, on an annual basis, writes billions of dollars of short-term loans entirely without human underwriters. This is the kind of technology that will be used in the digital bank as well, he adds.

“The good thing about an AI underwriter is that it is very consistent and you can curate it offline as well, and if you find there are biases, we can systematically fix the biases. Unlike AI, human beings, if you ask them to do something today, tomorrow they may do a different thing.”

When Foong was considering the offer to join the digital bank, there was also talk about bringing the forefront of AI to the bank and to the rest of the YTL Group. Eventually, they realised YTL was not the only entity sorely in need of AI solutions but almost every vertical across different industries, which led to the setting up of YTL’s AI Lab.

“The main difference between the AI revolution versus any other tech revolution, such as the internet or cloud, is that to build a regular system, you need a lot of talent to do it and some computation, whereas in AI the majority of the raw ingredient is computation and then you have some talent,” Foong explains.

In December 2023, YTL Power International Bhd, YTL’s utilities unit, announced a partnership with US tech giant Nvidia to develop AI infrastructure in Johor in a US$4.3 billion investment deal. The first phase was expected to be operational by mid-2024; however, the project saw some delays.

The Edge Malaysia reported that the delay in the project’s commissioning this year was due to the delay in the chipmaker’s production. In the latest update, Morgan Stanley said the situation had been resolved, with the first shipment expected by year-end.

The AI portion of YTL Power’s data centre project — fully funded by the company — was expected to house Nvidia H100 Tensor Core graphics processing units (GPUs), Nvidia’s top product then.

In March 2024, this was changed to the newer and much faster GB200 NVL72 chipsets, with a fund manager, who follows the sector, explaining, “For the same space, you get more processing power”.

This is YTL’s strategy to invest in computational power because without it, it is a non-starter. “Even if you have the talent, you are not going to be able to build AI. Now that we have this AI factory, this big AI supercomputer in Johor, what YTL AI Lab is doing is bridging the technology with its actual use cases,” Foong explains.

“The lab was set up to talk to governments, big companies and small and medium enterprises (SMEs) to help them use AI to build their own tools and solutions. And this includes potentially training LLMs based on local knowledge and languages.”

Ultimately, an AI factory embodies the industrialisation of AI development, providing the robust infrastructure needed to support the next generation of intelligent applications.

Taking into account that this is possibly the trend moving forward, Foong says there is no question that in the near future, we will see a lot of customised LLMs in the market being used for different cases, such as for securities, investments, banking and manufacturing, to name a few.

The first thing that needs to be developed is a standard or a Sirim for LLMs. “This is why we did research into it and built the first MalayMMLU (Massive Multi-task Language Understanding) and by building this facility, it will empower other institutions to build customised LLMs for Malaysia.”

YTL AI Labs built the MalayMMLU in collaboration with Universiti Malaya, which was released in September. The MMLU was recently accepted at the prestigious Empirical Methods in Natural Language Processing 2024 conference.

Ultimately, the MalayMMLU is a comprehensive benchmark designed to evaluate LLMs in Bahasa Melayu. By providing a standardised mechanism for evaluating LLMs across multiple-choice tasks, the MMLU benchmark facilitates fair and comprehensive assessments.

This benchmark is crucial for improving LLMs’ understanding of Bahasa Melayu and advancing applications in education, healthcare and public services, fostering culturally relevant AI solutions for Malaysia and Southeast Asia.

The first MMLU was a benchmark from the University of California, Berkeley and all the LLMs around the world are competing on that MMLU by answering a bunch of multiple-choice questions, predominantly in English.

It’s the same with the MalayMMLU, where there are 24,000 multiple-choice questions that different LLMs get to “take their exams” by answering 24,000 questions covering the entire F-12 spectrum across 22 subjects from the Malaysian education curriculum.

After which, the results are published in a table ranking these LLMs in Malay proficiency. For example, now, GPT-4 is ranked at the top and it’s one of the smartest LLMs around.

“Essentially, an app developer gets to choose, more accurately, which LLM is good for them. There are a lot of efforts right now to build applications using LLMs for customer services, depending on which customer segment you are targeting and if a lot of your customers tend to speak Malay, then you want to have an LLM that performed really well in Malay in customer service benchmark, specifically,” Foong says.

“There are a lot of projects, such as trying to make personal tutors, which means the LLMs used need to be benchmarked against standards for a subject, so if you are teaching maths and science, pick an LLM that scores well in maths and science.”

YTL AI Labs has also finetuned its own LLM and the lab is currently focused on refining it to provide the best value to those who need it.

“It’s always good to look at an LLM as an operating system. You generally don’t use the operating system to do your work, you use applications like Microsoft Word. So, think of it as the LLM is an operating system where a lot of other apps can live on and if the LLM is good enough, you almost don’t see it and just interact with the application itself.”

The lab is also actively pursuing a “forgetting technology” for the LLMs to forget certain information it has learnt in the event that the distribution of data within the LLMs parameters appears to be biased. This technology will allow the removal of targeted data to correct biases while at the same time safeguarding financial liabilities.

“In the world of LLMs, each time we train it, especially if it’s from the ground up, it can cost tens of millions of dollars. A factory reset becomes expensive when training with so much data and the outcome might not be guaranteed,” Foong explains.

“The technology is still nascent and a few countries are working on it, mostly among US and Chinese tech companies.”

Born, raised and educated in the suburbs of Kuala Lumpur until secondary school, Foong was especially interested in mathematics and physics. During his time at Sekolah Menengah Kepong Baru, he was in the first batch of students honing its Kelab Rekacipta dan Innovasi (Design and Innovation Club), a moment that shaped his life.

“We built a few things — silly things in hindsight — but that gave me the taste of building and inventing things. I remembered feeling that I wanted to do this for life, even though my dad wanted me to be a doctor,” shares Foong.

“The funny thing is, our family is very maths-oriented. I have three other siblings, two of whom ended up studying maths-oriented subjects. My sister went on to become a doctor and my dad finally got his wish.”

With the love, support and guidance of his family — on top of the fact that the sight of blood makes him queasy — he decided engineering was the way to go rather than medicine. Enrolling in Purdue University in 1995 was an “interesting juncture in time”, says Foong because it was completely unplanned.

It was the beginning of the internet revolution and Foong had a front-row seat. “It was a big experience for me because Purdue is located at the centre of the US, in West Lafayette, Indiana and all the internet cables travel through that place.

“I got to experience the internet boom firsthand and because it’s a research university, we had all sorts of, like, high-tech equipment and high-tech computing systems to play with. That pretty much sparked my interest in computing in general.”

Then, during one of the summers he spent in the US, he wanted to go on a trip but didn’t to fund the sojourn using his father’s money. So, he went on a hunt for the highest-paying job he could land, which turned out to be an AI research lab called SEAS Lab.

“I didn’t know what AI was at that time. But it was really interesting, so much so that when I graduated, my professor asked if I wanted to set up a company to continue to do this and that’s what we did. That is how I set up Simulex with a couple of my professors in 1999.”

His first company, Simulex Inc, was a boutique research firm specialising in AI and advanced analytics for predicting human behaviour. He started the company in 1999, having just graduating from Purdue University with a Bachelor of Engineering in electrical engineering and his work ranged from big-data distributed computation to social-sciences encoding into mathematical and algorithmic models.

It was his first foray into AI. “I told my mom about it and she didn’t know what I was talking about,” Foong quips, “but it was a really interesting project for predicting human behaviour, especially on ethnic conflicts and terrorism. It gave insights into how to apply computing to human behavioural activities.”

During his close to 14 years at Simulex, he consulted for various branches of the US Department of Defense and Fortune 500 companies, after which the company was sold to a major defence contractor.

After that Foong and his friends decided that they wanted to use predictive human technology in the banking space, thus founding MoneyLion Inc in 2013. “My two buddies used to work at the same research lab that I worked at and they used to be programmers but they moved to the investment banking space. They left their jobs to join me in starting MoneyLion together, with the intention of serving the middle-class Americans at that time as it was the tail end of the great financial crisis.”

As the CTO of Moneylion Inc, Foong led a team of over 300 engineers and data scientists to bring the forefront of AI into consumer finance, specifically to create the most scalable and high-frequency loan underwriting system examining thousands of variables of each applicant with effective use of machine learning algorithms.

The underwriting system systematically lowers the risk of the portfolio and provides equitable credit consistently. The same analytics prowess is also used to examine every financial data point to provide the most informed and accurate advice back to the consumer, fostering healthy, long-term financial health.

MoneyLion went public on the New York Stock Exchange on Sept 23, 2021, after merging with Fusion Acquisition Corp, with a valuation of US$2.4 billion. Among its leading investors are Blackrock, Edison Partners and Greenspring Associates.

Save by subscribing to us for your print and/or digital copy.

P/S: The Edge is also available on Apple's App Store and Android's Google Play.