GovTech launches LLM for data-driven policymaking in Singapore public sector

Oleh Si Ying Thian

Singapore public sector officers can use a simple chat interface to extract insights from government databases via SENSE; MOH first ministry to start using the application.

SENSE product team (left to right): UX designer Tan Lay Hui, UX designer Shaina Tan, Software Engineer Chadin Anuwattanaporn, Deputy Director Yeo Yong Kiat, Software Engineer Wilson Wan, Software Engineer Loh Wei Jun, and missing Software Engineer Ani Adhikary. Image: GovInsider

Policymaking is complex – and throwing hard numbers into the mix can make the process even more difficult, especially for those who do not specialise in data or technology.

 

But data continues to play a key role in driving informed decisions in public institutions, such as tracking and realigning policy interventions to achieve agency goals.

 

To make data more accessible for Singapore public service officers, a team at GovTech Singapore has developed a large-language model (LLM) application, known as SENSE.

 

Equipped with a natural language chat interface, SENSE allows officers to extract insights from government databases by simply asking the model for quantitative insights – as they would do so when interacting with a data analyst. 

 

Speaking to GovInsider, GovTech’s Deputy Director (Policy, Strategy & Development), Yeo Yong Kiat, says SENSE could potentially save up to two to three months of an officer’s time that is currently required to access data insights for a policy review.

SENSE LLM allows users to retrieve the meta data (left), and execute a code based on the user's request (right). Image: Yeo Yong Kiat
 

Policy officers typically go through the following process of retrieving data for a policy review: Start seeking approvals to access government databases, then liaise with data analysts to code the data that would facilitate the analysis, then go back and forth with them to refine the insights for the review.

 

SENSE also frees data analysts from the administrative burden of working through the nitty gritty of cleaning the data for the officers.

 

Analysts can then focus on what they specialise in – longitudinal studies and advanced analytics – that explore the impact of policy interventions over time, says Yeo.

 

To subscribe to the GovInsider bulletin click here.

Targeted approach to launch a whole-of-government product

 

For a start, GovTech is planning to onboard one government agency per month – and the first pilot is with the Ministry of Health (MOH).

 

The team has been working with MOH’s Healthcare Finance Division to train SENSE with the latter’s internal datasets, as well as put in place guardrails for data use and access.

 

From this month, GovTech has deployed a dedicated SENSE instance for use by the whole of MOH.

 

“I think it pays to have that flip side of innovation: To be slow, receptive to changes, and be a little more cautious, rather than go full steam,” says Yeo, highlighting change management and feedback loop as key factors to manage.

 

A key lesson he has picked up from the onboarding process is what he calls “creative tension.”

 
Ministry of Health (MOH) is the first agency to be onboarded to use GovTech's SENSE. The team is hoping to achieve an "ecosystem effect" by targeting health and social service agencies next. Image: MOH

“You have a party pushing for the product, and another who gatekeeps very strongly. The end solution is not so much a compromise, but the process helps the solution to be better,” he notes.

 

And why start with one of the most sensitive sectors? Mainly due to Yeo’s familiarity with the processes at MOH as the previous Assistant Director of Healthcare Finance.

 

The other reason is the team’s aim to achieve an “ecosystem effect” with Healthier SG, the national strategy focusing on preventative care.

 

Instead of moving from one agency to another without a goal, the team wants to reap the synergies between the health and social sectors.

 

After MOH, GovTech is planning to work with the Health Sciences Authority (HSA), National Healthcare Group (NHG), Ministry of Social and Family Development (MSF), and Ministry of Education (MOE).

 

“Targeting these agencies creates a virtuous cycle,” he explains, adding that health and social service agencies have an incentive to share data with each other for the benefit of their clients – regular citizens, by and large.

 

To subscribe to the GovInsider bulletin click here.

Enabling a culture of reuse, no duplication

 

SENSE only deals with structured datasets, which refer to data that conform to a defined format or structure.

 

“We believe in reusability and want to find other products that have synergy with us. Then we can call an API, rather than build it ourselves,” Yeo explains.

 

Application programming interfaces (APIs) allow apps to interact with one another. An API call allows one app to request data or services from another.

 

Yeo shares that his team was careful not to duplicate efforts, and direct the limited resources and attention to solve a specific problem instead.

 

“I found a bunch of people here who are very interested in problem solving. They're not in love with the solution, they're in love with the problem,” he says.

 

Pair, developed by GovTech’s experimental tech arm, Open Government Products (OGP), is another LLM contextualised for the public sector and works with unstructured data. Pair has since garnered over 11,000 users across more than 100 government agencies.

‘End to end’ data sharing the next step

 

“Here at GovTech, we believe that if you want to solve a problem, you break it down into smaller parts and tackle those that are easier to solve first,” he says.

 

Right now, SENSE only works for data access and analytics, but the longer-term plan is to facilitate inter-agency data sharing across the whole-of-government.

 
Yeo sharing about SENSE at the National Healthcare Group (NHG)'s Singapore Healthcare & Biomedical Conference (SHBC). Image: Yeo's LinkedIn

“Once you touch data sharing, you have to go into multiple agencies at the same time – and it becomes a wicked problem,” he explains, adding that over time multiple agencies may become more receptive to use SENSE for data sharing once they are familiar with the product and processes.

 

User access control will be key in enabling data sharing, which Yeo explains is the ability to control different users’ access to levels of data.

 

Currently, user access control is manually coded for each agency, making it challenging to scale. “We want to eventually give this power to the agencies themselves to decide user access,” he says.

 

Yeo shares that the team is also exploring open-source LLMs to “future proof” government applications like SENSE, which is currently built on OpenAI with a zero data retention clause.

 

He highlights inter-agency data sharing as one of the unique needs of the public sector, which may not always be served by private offerings.

 

On his team’s aspirations for SENSE, he says: “We want to evolve it into what Apple calls the next layer of AI apps, where AI is not the main feature... but (is) meant to solve a bigger problem.

 

“We intend for it to be an end-to-end platform that allows government officers to freely share, explore, extract data analysis – without having to go through too much coordination, spending too much time and having to pick up very technical expertise.”

 

Yeo and his team demonstrated the use of SENSE on open datasets from HDB at the Singapore Public Service Week Learning Festival 2024. To find out more about how SENSE works, you can watch their video here.