What happens when you give frontline workers a government AI platform to build on

In the middle of a shift, a pediatric nurse needs to check the internal database and verify a drug dose before it reaches a child.  

The calculation is weight-based, where every milligram counts. 

For a long time, nurses at KK Women’s and Children’s Hospital (KKH), a public pediatric hospital Singapore, navigated this the slow way: Go to intranet, search the document, manually calculate each drug dose, and double check. 

Meanwhile, a junior nurse has questions about ward protocols, which require another search for a document buried in the system.  

The questions get answered, forgotten, then asked again. 

Now, the nurses can open just one browser tab to access the “Pediatric Nursing Sidekick” platform on their ward laptops to tackle both problems.  

A chatbot handles the protocol questions, while a calculator takes a child’s weight. 

This has cut the time spent to calculate the doses from minutes to seconds, and flag in red to the users if the doses run too high, says Advanced Practice Nurse (APN) Joanne Cheng, Assistant Director, Division of Nursing at KKH, to GovInsider. 

Two months in, the chatbot logged over 200 conversations within the hospital alone. The pharmacy division has also since approached the team to add their own data. A SingHealth cluster-wide rollout is now under discussion as well. 

The nursing team behind this shares more about the lesson takeaways. 

Government AI tool turned citizen developer environment 

Before the chatbot landed on Pair, which is the Singapore government-secure GPT tool, the team first tried Telegram’s low-code option to build a bot, but hit a wall to scale its use due to data protection rules and device restrictions.   

A collaboration with final-year students from Republic Polytechnic came next and while it produced something workable, the team did not have the technical background to maintain it. 

Building the tool, it turned out, was only half the problem. The other half to sustain the tool was to be able to maintain it themselves. 

When Pair became available to the public healthcare sector, it removed the obstacles the team had hit before: no server to maintain, no budget to approve, no patient data to worry about, and no device restrictions. 

The nurses could access Pair on their ward laptops, update “Pediatric Nursing Sidekick” chatbot themselves, and keep it running without depending on anyone else. 

KKH’s Assistant Nurse Clinician, APN Lai Liling, shares that her team would prompt Pair in “layman terms” to create a Python code to create the chatbot, and iterate the code until they were satisfied with the first prototype. 

“Pair was quite easy to use. In the sense, there were instructions on how to start using the chatbot and load things into it according to what we want,” she explains.  

In the end, it wasn’t a structured training programme or a top-down mandate to innovate, but simply having access to a government AI platform that led the nurses to experiment and build tools to solve their own problems.  

Widespread adoption as the main driver 

Lai is candid about the challenges faced by the team at the start: “a lot of times we wanted to give up.” 

“Along the way, we managed to create something on our own and it works, there’s this sense of satisfaction,” she adds. 

The chatbot was quickly adopted by the nurses, Cheng notes, as it directly tackles their pain points and is easy to use on their ward laptops. 

The high adoption was encouraging for the team, who then used the opportunity to continuously test it with nurses and incorporate their feedback into the chatbot, says Assistant Nurse Clinician, APN Zhu Yu. 

When nurses started requesting refinements, like specific information to be included, it signalled genuine adoption rather than passive use, she notes. 

This organic word-of-mouth and usefulness were what drew the pharmacy division in and caught the attention of cluster leadership to consider scaling it. 

Hackathon as the accelerator 

The continuous iteration meant that the tool was already well-fitted to real needs by the time it reaches a wider audience.  

The nursing hackathon, as part of the annual SingHealth’s Nursing & Research Innovation event, came up at around the same time the team was working on their chatbot. 

Cheng shares that besides providing a platform to present the chatbot to a wider audience, the team benefited from the one-day workshop to refine their problem statement, hone their Pair prompts, and pitch their idea to get stakeholder buy-in. 

With seven minutes to pitch the chatbot to a wider and more diverse audience, the team made it to the top ten. 

The value of ‘networking widely’ 

“It's really a lot of talking to many different people, and then people link you to different people, and then it led us to where we are today,” says Assistant Nurse Clinician, APN Regina Lim. 

Networking for the team was less over facilitated channels but more informal outreach, asking anyone who might know something and following every thread. 

"Really go and speak to everyone. Anyone. You will be surprised. Some people have resources you may never think of,” says Lai. 

Notably, what made the networking effective was also the team’s openness to partial solutions. The team discovered early that many people around them had started something similar but never finished. 

“That's where you work with the team together. You share your idea, I share my idea, something better comes out,” says Cheng. 

Every failed attempt, Lai reflects, provided a clearer sense of knowing what they need to build and improve the chatbot. 

And perhaps the takeaway for civil servants sitting on an unsolved problem is that the expertise they need may already exist somewhere in their network.

Gov in 60 Seconds (-) - Moving from vibe code to real impact