The SIT team, comprising (from left to right) Nah Yong En, Dr Budianto Tandianus and Isaac Pek, with the Guest-of-Honour, Senior Minister of State Tan Kiat How, at the National AI Prompt Design Challenge 2024. (SIT Photo: Isaac Pek) 

Misaligned project scopes, inappropriate budgets, and timelines often lead to researchers being rejected when applying for grants. But what if there was a tool that could work out these inconsistencies? That’s what Dr Budianto Tandianus and his team members, Mr Isaac Pek and Mr Nah Yong En, envisioned. The trio brought this idea to life when they represented the Singapore Institute of Technology (SIT) at the National AI Prompt Design Challenge 2024, held from 10 to 11 September.

“Most times, when we want to write a research grant proposal, (research) agencies have in mind things they want to work on. As researchers, we also have ideas we want to carry out,” said Dr Budianto, a Senior Professional Officer with SIT. 

He noted that this conflict of interest often led to rejected proposals. Thus, the team decided to explore the creation of chatbots to address several challenges identified during the research process, particularly the mismatch between the scope of research proposals that grant funding agencies seek and those submitted by researchers. Aligning these proposals with the funding agencies' expectations could save significant time and effort. Even proposals that are rejected could be repurposed and resubmitted to other grant opportunities. This is just one potential use of the chatbot; it can also assist new principal investigators with tasks like budget planning, resource allocation, and timeline management.

The team eventually submitted three bots, including a project management and question analyser chatbot, and came in second out of more than 100 teams with their research grant proposal analysis bot, named Reviewer #2.

Built to help researchers obtain grants, the tool aligns their proposals with grant agency requirements.

By simply uploading their proposal and the grant’s guidelines on the bot, the proposal can be scored across four components: project alignment, proposal review, project timeline, and budget.

The researcher can then chat with the bot about improving component scores.

Reviewer #2 is able to provide an analysis of a grant proposal and check for alignment with a grant call. (SIT Photo: Isaac Pek)

An example of Reviewer #2 conducting a proposal budget review; adjusting the budget to fit the grant call’s funding quantum. (SIT Photo: Isaac Pek)

“The whole objective of this research grant proposal analyser is to allow you to take your rejected proposal and find other grants it can fit into,” said Isaac, Senior Manager at SIT’s Centre for Digital Enablement (CoDE).

However, its use case is not limited to rejected proposals. It can also come in handy at the drafting stage by suggesting which areas to change as researchers write up their proposals, he added. The bot’s name is a tongue-in-cheek reference to a common belief in the research community that the second reviewer always gives the harshest feedback, said Dr Budianto. 

On Level 5

Their project started in August 2024 when a Facebook advertisement for the challenge caught Dr Budianto’s eye.  Intrigued, he roped in Isaac. However, they needed three people to form a team.

Isaac, who had previously attended a generative AI (GenAI) training workshop conducted by Yong En, thought the latter would complete the line-up. “I knew that he is at the forefront of GenAI initiatives within SIT’s Teaching and Learning Academy (STLA),” Isaac said.

Isaac also saw Yong En’s teaching background as a valuable asset, believing his perspective would bring a fresh approach to the challenge. The educational technologist, who helps SIT academic staff integrate technology to improve students’ learning experiences, agreed to join the team.

Yong En explained, “We decided that we wanted to make chatbots that helped with research…we knew we had a list of pain points of the research process, and we wanted to see where it could be useful.”

SIT Team “Level 5”, comprising Yong En, Dr. Budianto, and Isaac (from left to right), at Level 5 of their office. (Photo: Isaac Pek)

Input, Output, Workflow

Although the team had basic knowledge and understanding of GenAI, they were not experts in the field. The short competition time frame of two days to complete their bots made the working process for Reviewer #2 even more challenging.

The biggest challenge was ensuring that the chatbot could hold a proper conversation with the end user. To achieve that, the team needed a set of good system prompts—instructions on how the chatbot should behave. If the set was not secure, users could instruct the bot to do things outside its core scope, like sing a song or say something offensive.

“This set of instructions not only guides the chatbot but also ensures that your chatbot is properly secured, preventing instructions from being leaked to the end user,” explained Isaac.

“We prevented this by adding instructions at the system prompt level to ensure that our instructions aren’t divulged,” he added.

While ensuring accurate output was important, so was the quality of the input. As the team could not use actual research grant proposals to test the bot, Dr Budianto drew up dummy drafts instead. “I spent a lot of time creating the demo documents we used to validate the chatbot's functionality.”

The team also had to ensure that users could communicate with the bot seamlessly, and a workflow was necessary to do so.

This was where Yong En’s experience in creating lesson plans came in handy. “There’s a sequence of things. So, I would sit down and think, what’s the sequence we want?” Yong En shared.

The competition experience has inspired the team to consider hosting an AI prompt design challenge for academic staff and students within SIT. This will allow the SIT community to gain a deeper appreciation of GenAI.

But above all, they are glad to have done SIT proud. “Out of the five winners of the challenge, we’re the only institute of higher learning,” Dr Budianto beamed with pride.

Congratulations once again, Team “Level Five!” (Photo: Isaac Pek)