Singapore General Hospital staff are discovering the ease of creating AI bots to help their work. They created 48 AI assistants that battled it out at the very first AI bot contest in SGH.
Many of us may have heard of or tried using ChatGPT, an AI chatbot that can answer questions, retrieve information and generate content for you based on your prompts. For us in SGH, we have Pair, an AI tool developed by
Open Government Products (OGP). It is a secure version of ChatGPT for Singapore public servants, including public healthcare staff. Many colleagues have found that they can use Pair to create their own chatbots without having to learn any coding skills.
The inaugural
Singapore General Hospital (SGH)
Pair-a-thon: AI Assistant Contest drew 48 teams, involving 150 participants from 22 departments, who had over a month to build, test and refine their AI bots. Six outstanding teams were then shortlisted to deliver a five-minute pitch to the judging panel at the finals on 3 March 2025 and demonstrate the value of their AI bots.
No code AI solutions
The Most Impactful Bot award went to the Nursing SOS Library AI assistant, which serves as an intelligent interface to the relevant policies stored in SGH Infopedia (SGH’s intranet that serves as a staff resource library). Nurses, who form almost half of our workforce, can now input nursing-related questions and get step-by-step instructions from the bot, based on policies and protocols.
The team of Staff Nurse Reina Cheong, Senior Staff Nurse Lee Jian Ting, Nurse Clinician Dezarae Ang and Nurse Clinician Tricia Lang wanted to "get rid of silly stuff", to eliminate workflow inefficiencies while improving both nursing productivity and patient care delivery.
Jian Ting and Dezarae had not used Pair prior to the competition but didn’t have trouble picking it up. “I just typed in instructions like I was talking to a person and uploaded data and protocols. Pair makes work easier by processing data much faster than we can manually, and it gives quick answers instead of us having to search through protocol documents or rely on colleagues,” shared Jian Ting.
/nursingSOSlibgp400.jpg)
Most impactful Bot: Nursing SOS Library
They are now obtaining the necessary approvals to officially launch this AI tool for nurses. In the meantime, the team is analysing which policies are most frequently referenced and which, though rarely accessed, remain critical. They will prioritize these to incorporate into their Pair Assistant.
In addition, the team is also exploring robotic process automation (RPA) to help keep the information up to date, be it the policies and procedures or contact information. They will also continue collaborating with phone directory staff to manage the e-Phone Directory Pair Assistant with RPA.
Most Production-Ready Bot i.e. most ready to be rolled out for use: X-Ray Personal Assistant (XPA)
SGH radiographers worked along a similar vein of developing a solution that centralises information on a single platform. This dramatically reduces the time radiographers spend searching for protocols and guidelines, making their workflow more efficient.
Our radiographers have the unique challenge of working across a wide range of settings - Polyclinics, Outpatient, Inpatient, and Emergency Departments.
“We realised that the constant rotation between these areas led to inconsistencies in workflow execution, increased error risk, and difficulties in accessing up-to-date protocols, which were scattered across multiple platforms like Infopedia and WhatsApp,” shared Crystal Chin, a member of the Pair-athon team that includes Celine Tan, Crystal Chin and Eugene Tan.
“We were new to Pair before the competition, but we quickly got up to speed by studying the Pairwise guide and attending an orientation session conducted by the contest organisers,” said Crystal.
One of the biggest challenges was ensuring the chatbot’s accuracy and reliability—particularly minimising “hallucinations” and incorrect responses. To overcome this, the team implemented features such as confidence ratings, citations, relevant URLs, and a user-friendly interface.
/XPAteam400.jpg)
Celine Tan, Crystal Chin and Eugene Tan who came up with the X-Ray Personal Assistant (XPA) AI tool
The most rewarding part for the team was witnessing how these solutions enhanced user experience and how AI could be fine-tuned to enhance clinical practice, especially to support better decision making during high pressure situations.
Feedback from a junior radiographer attesting to XPA’s ease of use, “XPA is very user friendly and efficient! I didn’t have to craft questions in full sentences, and I like how it provided additional resources from Continuing Professional Development (CPD) Lectures that helped in answering my questions.”
The team plans to expand their solution to address all imaging modalities challenges, transforming it into a comprehensive educational resource for radiographers. As AI continues to evolve, the team sees immense potential in refining and expanding XPA to serve as a robust, indispensable tool in radiography.
Most Innovative Bot: OncoNarratus
The creators of OncoNarratus took an unconventional approach, harnessing AI for education to teach – wait for it – empathy!
OncoNarratus is a game for empathy education. Users can create their own character, go through a “play-your-own-adventure” format where they are faced with patient-fronting scenarios, and complete a quiz at the end. Users will then receive feedback on what they did well, and how they can improve their empathy skills. The bot utilises large language model (LLM) hallucination to generate adaptive, varied scenarios.
The project brought together oncologist
Assistant Professor Jolene Wong, research officer Kevin Fo and Associate Professor Kevin Yap, Pharmacy Practice Manager who is also Serious Games Lead in the Division of Pharmacy. They were in the beginning stages of ideating how to train clinicians through innovative technology like generative AI, so the timing of the competition was perfect.
“We identified a challenge in healthcare education where traditional teaching methods like lectures, role-plays, and case studies are still predominantly used to develop narrative competence and empathic responding in young clinicians. We saw an opportunity to use generative AI to create a novel approach to this challenge by addressing what we call the 'Three-body problem' in healthcare education - merging the humanities (empathic responding and narrative competence), science (medical knowledge), and technology (digital health technologies) into an effective educational solution,” explained Kevin Yap.
/onconarratusteam400.jpg)
The two Kevins presenting during the Finals.
One major hurdle was optimizing their assistant to create engaging user interactions – for example, they faced difficulties converting tables and figures from journal articles into formats that Pair could effectively interpret. The team had to spend considerable time ensuring prompt changes were effective and had to think creatively to achieve specific behaviours.
Despite these challenges, the process was rewarding. “We were motivated by the opportunity to showcase Pair Assistant's potential beyond typical operational and administrative workflows. The best part was during the evaluation phase, where we received positive feedback from users, which was particularly touching and validated our efforts,” said Kevin Fo.
Both Kevins had prior experience in prompt engineering, which made learning to use Pair relatively straightforward, though optimising OncoNarratus still required considerable trial and error.
“Pair has great potential, particularly for education - it could be an interesting tool that translates across various educational domains and specialties. From personal experience, Pair boosts efficiency in daily work tasks,” shared Kevin Fo.
The team likes Pair as it’s user-friendly, especially since most users already have experience with AI chatbots. They also like that Pair is freely available to public hospital employees and is powered by a top-performing LLM. The document upload feature has also proven to be especially useful, which they took advantage of to create OncoNarratus.
“We aim to enhance the bot by expanding its scope to cover a wider range of serious illnesses with the goal of establishing a generative AI-based educational pedagogy that could be implemented both nationally and internationally. Abstracts* of our Pair-a-thon results have been accepted for the upcoming Multinational Association of Supportive Care in Cancer Annual Meeting in the USA and the Edulearn International Conference on Education and New Learning Technologies in Spain. We plan to publish our data once we gather more results,” said Kevin Yap.
Encouraging bottom-up innovation and experimentation
Jonathan Tan from AI and Automation, Future Health System, spoke for the organisers.
“While there has been a lot of excitement about Generative AI, there is still a need to translate this into practical value at an organisational level. We believe the best way to do so is through encouraging bottom-up projects, as teams across the hospital experiment and discover best ways to apply this technology to their work. Such bottom-up innovation is now possible thanks to tools like Pair, which allow ground staff to create their own AI solutions without coding expertise. We organised the SGH Pair-a-thon as a platform to encourage this and to recognize teams for their pioneering efforts,” explained Jonathan.
Many teams shared with the organisers that the contest helped them to gain firsthand experience of the capabilities and limitations of LLMs, and to understand the importance of proper evaluation before deployment.
“Apart from improving SGH’s AI literacy, it was also important that all of us (organisers and participants) had fun – and I think we did!” observed Jonathan. “We were also really impressed by the effort that the participating teams put into developing and testing their Assistants within a short timeframe,” he added.
Jonathan will be monitoring usage of the bots submitted for the contest to determine the winner of the Sustained Use Award.
We love mail! Drop us a note at lighternotes@sgh.com.sg to tell us what you like or didn’t like about this story, and what you would like to see more of in LighterNotes.
