Red Hat has set up a centre of excellence (CoE) for artificial intelligence (AI) in Singapore, stationing forward-deployed engineers (FDEs) in the city-state to help organisations move AI projects beyond the proof-of-concept (PoC) stage and into production.
Supported by the Singapore Economic Development Board (EDB), the CoE will help Red Hat’s engineering teams better understand what customers in the region need.
“We have a roadmap of the functionality and projects that are coming downstream to be generally available, but customers sometimes differ in terms of their priorities,” said Daniel Aw, vice-president and general manager for Asia-Pacific at Red Hat, at a media briefing on the sidelines of Red Hat Summit: Connect in Singapore.
“What this AI CoE will do is to bring the engineering capability locally here, and we also want to reduce what we call the feedback loop back to headquarters, back to engineering, on the features that customers in Asia-Pacific are looking for,” he added.
The CoE will launch with around 18 people, including FDEs, architects and project managers. “We’re starting small,” Aw said, adding that the headcount could grow depending on demand and the projects the team takes on.
Red Hat does not have a fixed set of criteria for sending FDEs to customer environments. Its pre-sales and specialist sales teams engage customers first, and the CoE takes over once a project reaches a certain stage.
“A lot of times we realise, as we work with customers, they have PoCs, and then the next step sometimes takes a long time, and the specific requirements may not be able to scale immediately. That’s where the CoE then comes in,” Aw said.
That could mean integrating software that Red Hat doesn’t yet support, including tools built by Singapore startups or other companies in the region – though any project must also align with Red Hat’s platform strategy.
The CoE’s engineers will build reusable software architectures for common industry problems and contribute code to upstream open source projects, targeting enterprise requirements such as system deployment, runtime optimisation and platform management, according to Red Hat.
“Red Hat is really in the platform business,” said Aw, describing a foundation on which customers can build reusable AI agents, create identity for every AI agent, as well as bake in AI governance and safety.
Yueh Chinn Seah, vice-president of Digital Industry Singapore, said the roles the CoE will create are good jobs that Singaporeans want. They also support the National AI Strategy’s goal of tripling the country’s pool of AI practitioners to 15,000, she said.
There are already over 100 AI CoEs in Singapore, but Red Hat’s focus on production will help make this one stand out, Seah noted. “Red Hat is different from other companies. It’s always about production readiness,” she said.
Seah sees that expertise benefiting multiple industries. For instance, an engineer who works with Red Hat to automate workflows at a hospital could then go on to deploy similar automation at a bank or on a factory floor. “That is how we foresee AI going to be proliferated across the economy, across different sectors, across different companies,” she said.
Helping customers build sovereign AI capabilities is also part of the CoE’s remit. Most organisations consume AI by sending requests to frontier models through application programming interfaces (APIs), Aw said.
Red Hat’s main pitch, he added, is to turn them from token consumers into token providers that host their own AI models – especially open source models with open weights – within their own environments.
Working with DSTA
Singapore’s Defence Science and Technology Agency (DSTA) signed a memorandum of understanding (MoU) with Red Hat on the morning of the briefing to work through the CoE on projects the agency wants to prioritise, building on two years of collaboration.
The two organisations will collaborate on AI, cloud computing, application modernisation and edge computing capabilities, as well as DevSecOps and day-two operations practices for on-premise and off-premise systems.
DSTA is building safeguards against the biggest risks posed by AI agents into its AI platform “upfront and not bolted on after”, according to Akash Philip, deputy director of AI development at the agency.
Philip pointed to the lethal trifecta, a concept from the AI security community that describes AI agents with access to sensitive data, exposure to untrusted content and the ability to communicate or exfiltrate data out of a network.
In DSTA’s case, agents have access to sensitive data all the time. Agents may need all three elements to be useful, but the combination is a recipe for disaster, Philip said.
But AI risk, he said, “cannot be a crippler”, even though the same three elements were behind the July incident in which OpenAI agents escaped their test sandbox and breached Hugging Face’s systems. “You still want to be able to leverage some of these technologies for the betterment of our customers.”
DSTA keeps its own AI experiments “controlled and not like free fall”, Philip said, scrutinising, through machine learning operations (MLOps), what agents can access and what permissions they hold.
“It is not a case in which you can just take a model out from Hugging Face and then just deploy it into production and expect it to work,” he said.
Teams must pick the right models, test them against the problem at hand and account for security before moving them into production. DSTA puts models through that process before they reach its on-premise and closed networks, freeing development teams to concentrate on operational use cases.
Philip said working with Red Hat combines its expertise in open source and scaling AI initiatives with that of DSTA engineers who understand defence and on-premise requirements.
“Our considerations are absolutely unique, but not the easy ones. They’re quite challenging simply because of the diversity within our environment,” he said. With the field moving so fast, “it is impossible to do everything yourself”.
Leave a Reply