If you ask a chatbot why some parts of Singapore feel hotter than others, it will almost always give you a confident answer. Often we don’t get a reliable explanation of where the answer came from.
The National University of Singapore (NUS) says it aims to fill this gap with AI Sense Maker, an interactive research tool that will be launched on August 20 to coincide with the start of the new academic year.
Built in-house over 12 months by NUS Libraries and NUS Information Technology (NUS IT), the platform allows students and academics to enter a question in everyday language and returns an overview with citations, a set of related concepts, and a knowledge graph showing how those concepts are connected.
The answer is extracted from two NUS repositories: ScholarBank and Digital Gems. ScholarBank houses selected digitized papers dating back to the 1950s, as well as the university’s original research output, including journal articles, conference papers, and dissertations. Digital Gems houses digitized rare materials focused on Southeast Asia, some of which date back to the 17th century. Between the two, approximately 150,000 items will be supplied at launch.
According to NUS, the university is the first in the region to build this kind of tool in its own collection.
Tan Shui-Min, chief information technology officer at NUS, said it typically takes researchers about two weeks to complete the initial stages of research: finding relevant papers, identifying relevant topics, and even tracing citations to other papers. “With AI Sense Maker, we can cut that down to just 30 minutes or even less,” she said at a media briefing earlier this week.
For NUS University Librarian Natalie Pan, sourcing issues are more important than time savings. She noted that students and researchers who go directly to chatbots for answers skip the time-consuming tasks of collecting, weighing, and synthesizing materials, but often don’t know what the answers are based on.
“Many AI tools are not designed with provenance in mind,” Pan says. “We often see the answer, but we don’t know what’s driving it – what sources, what kind of work it is.
“Researchers today don’t lack information; they lack time and clarity,” she added. “AI Sense Maker is designed to help people understand information, not just find it. Rather than replace critical thinking, it helps people ask better questions, discover new connections, and generate fresh insights.”
The platform uses OpenAI’s Large-Scale Language Model (LLM) and Search Augmented Generation (RAG) techniques to answer queries from ScholarBank and Digital Gems materials, rather than what the model absorbed during training. Questions not included in the collection will be rejected.
Magdeline Ng, deputy university librarian and cluster head for digital strategy and innovation, said the team used the Wikidata taxonomy when building the knowledge graph. However, this taxonomy was too granular to work with, so the team trimmed it to match what the NUS repository actually holds.
“The team had to find a balance between what they knew about the collection and what classification levels they should include,” she said. The resulting ontology is proprietary to NUS.
There are limitations at startup. The tool currently only processes English, with other languages in the pipeline as well. Also, while follow-up questions cannot be taken yet, Tan said these questions are scheduled for the next enhancement, which NUS IT plans on a three-to-six-month cycle.
An even bigger omission is the content of commercial magazines. Subscription databases for publishers such as Elsevier, Springer Nature, and Oxford University Press are external to the system. “We have not included them because we have licensing agreements with major publishers,” Pang said.
That said, ScholarBank holds a significant number of accepted manuscripts that are substantially closer to the published paper, with author versions finalized after peer review and before typesetting or copy editing by the publisher.
The knowledge base is expected to grow to 250,000 items and NUS IT is considering whether to host its own model rather than pay for a commercial model. Token consumption is expected to be manageable, “but we are definitely looking at hosting our own LLM in the future,” Tan said.
There have been internal discussions about adding collections from other universities, but no proposals have been made by any universities. Obtaining a license is one hurdle, Pan said. Another is that partner libraries have to make significant investments in preparing their own materials, and many libraries in Southeast Asia are still in the early stages of digitizing their collections.
AI Sense Maker is part of Libraries of Fantastic Things, an initiative by NUS Libraries that has also created a systematic review agent that teaches students how to review and synthesize existing research using library-specific videos and guides.
