When researchers at the National University of Singapore developed AiSee, a wearable device that assists visually impaired individuals navigate their surroundings, they turned to Meta’s open-source Llama model to power the breakthrough. The project illustrates how open-source artificial intelligence (AI) is evolving from a developer tool into a catalyst for societal transformation across Southeast Asia.
Open-source AI also democratises access to advanced AI capabilities, delivering measurable gains in cost savings, productivity, and innovation to organisations of all sizes. For example, the Singapore Academy of Law (SAL) and the Infocomm Media Development Authority (IMDA) rolled out a Llama-powered tool that can summarise legal judgments in 10 minutes, a task that previously took days. Those pre-generated summaries are available on SAL’s LawNet platform, which is used by roughly 75% of lawyers in Singapore.
"What we consistently hear from developers is that the transparency, customisability, and security of open-source models, like Llama, enable new levels of creativity and innovation," says Dr Rafael Frankel, Meta's director of Public Policy for Southeast Asia. Since its launch in 2023, Llama has been downloaded 1.2 billion times and is being used by enterprises, start-ups, non-profits, creators, scientists and public institutions.
Enabling trustworthy AI applications
AI has long been criticised as a black box or a powerful system that generates outputs but offers little insight into how it reaches conclusions. That opacity has slowed adoption, particularly in the public sector and regulated industries.
Open-source AI models, such as Llama, offer a way forward by providing transparency.
Open-source AI can make models more secure because the transparency inherent in open source allows for a wider community of external scrutiny that can identify and fix potential issues.Dr Rafael Frankel, director of Public Policy for Southeast Asia, Meta
Data control is another advantage. Open-source AI models, says Frankel, can be deployed locally, keeping sensitive data secure and compliant with privacy regulations, unlike proprietary models that often require sharing data externally. They also offer customisation, enabling organisations to meet specific regulatory needs. This is especially key for organisations operating across Southeast Asia, where regulatory frameworks vary by country.
Meta has built additional safeguards into Llama to strike a balance between openness and safety. “Before releasing any model, we conduct thorough risk assessments, extensive testing, and red teaming exercises. This helps us identify and mitigate risks proactively, ensuring that the models perform safely in real-world scenarios,” shares Frankel.
Privacy, he adds, is “front and centre”, with Meta carefully curating the data used to train Llama and excluding sensitive personal information. The company also provides application developers with safety tools such as Llama Guard and Prompt Guard, which Frankel describes as “safety nets to prevent misuse and ensure outputs align with ethical guidelines.”
Apart from that, Meta offers resources such as the Developer Use Guide, which outlines best practices and considerations for application developers, as well as mitigation strategies and resources to address risks at various points in the system. It has also worked closely with government agencies, including IMDA and Singapore’s Government Technology Agency (GovTech), to understand local priorities and requirements for safe and trusted AI.
Ultimately, AI safety is a shared responsibility. “We encourage collaboration with governments, civil society, and industry to promote transparency, accountability, and ethical standards to advance trustworthy AI for everyone,” says Frankel.
Turning demos into deployment
To spur the development of practical AI applications, Meta launched the Llama Incubator Program in Singapore in March 2025 with partners including the Ministry of Digital Development and Information (MDDI), IMDA, GovTech, Digital Industry Singapore (DISG), Enterprise Singapore, SGInnovate and AI Singapore.
“This six-month programme takes a multi-faceted approach to nurture and scale impactful AI projects, including building capacity, sharing best practices, and facilitating access to ecosystem partners and resources,” says Frankel.
The incubator provides start-ups and small and medium-sized enterprises (SMEs) with technical and business mentorship from industry experts to help them build and deploy AI-powered solutions. To ensure safety is built in from the start, participants will also gain hands-on experience with Llama’s trust and safety tools, IMDA’s safety testing framework, and AI Verify’s open-source testing toolkit, Project Moonshot.
On Oct 15, the programme will culminate in a Demo Day. Up to 40 teams — consisting of start-ups, SMEs, and public sector groups — will showcase real-world applications of Llama-based open-source AI. “To ensure promising projects can scale and continue delivering value, Meta is collaborating closely with local partners (including government agencies and ecosystem partners) to connect participants with potential customers and investors, as well as provide information and access to possible grants and other networks,” says Frankel.
He adds that the government track of the incubator also serves as a platform for the public service to adapt and scale AI solutions for its own needs, aligning the programme with Singapore’s broader ambition to lead in safe and trusted AI.
The Llama Incubator Program is among Meta’s initiatives that support Singapore’s Smart Nation 2.0 vision, which aims to empower local SMEs and attract start-ups from the region to build and innovate in Singapore. Meta has also expanded its multi-year Upskill with Meta programme to include AI-driven components. This will give SMEs and students hands-on experience with AI-powered tools, thereby cultivating a more skilled and productive workforce.
“We see ourselves as an ecosystem partner, working closely with government, academia, industry, and the developer community to advance AI capabilities and ensure that the benefits of this technology are widely accessible. We believe Llama’s open and accessible approach will empower businesses and consumers to build transparent, adaptable, and secure AI solutions that drive innovation and progress, while placing citizens at the forefront of technological advancements,” says Frankel.
