NUS students explore AI’s potential across space, business and healthcare

NUS students explore AI's potential across space, business and healthcare

An error occurred while preparing your download

From disaster response to patient safety, a group of award-winning NUS students are putting Artificial Intelligence (AI) to work on problems that matter. NUS News highlights how the winners of NUS' Outstanding Undergraduate Researcher Prize (OURP) are applying AI to address real-world challenges. AI's potential is best realised when paired with critical thinking, rigorous evaluation and thoughtful governance.

Galassia 5: AI-powered earth observation satellite

Most satellites transmit everything they capture back to earth, and this is a slow, bandwidth-heavy process that limits how quickly useful data can reach people who need it. A team of students from the College of Design and Engineering at NUS set out to change that.

An error occurred while preparing your download

Final-year students Tristan Voon, Grace Ng and Victor Soh, together with NUS alumni Lim Tze Yee, Hang Jin Guang, Lee Wen Da, Phyo Thu Yein Kyaw and Pan Yong Jing from the Class of 2026, built Galassia 5, an earth observation satellite that processes data onboard in real time. At its core is a neural processing unit (NPU), a specialised hardware that runs artificial intelligence (AI) and machine learning workloads directly on the device without relying on a ground connection. Combined with direct-to-user downlinking, which sends data directly from orbit to end users, Galassia 5 significantly reduced the time taken for critical insights to reach users who need it most. In disaster situations like earthquakes, faster satellite data transmission means less time needed for first responders to reach earthquake survivors.

An error occurred while preparing your download

Stress-testing AI as a research tool

Can AI tools be trusted for serious research? A team from NUS Business School decided to find out.

Final-year students Joshua Arjanto and Muhammad Khaizuran Bin Mohamad Rosle, as well as NUS Class of 2026 graduates Hu Shiyou, Chen Xiaojing (Crystal) and Asher Ethan Koh - put leading AI research tools, including OpenAI Deep Research, Google Gemini Deep Research and DeepSeek, through their paces. The task was to analyse the business impact of emerging technologies, such as AI, quantum computing, robotics, space technology and fusion energy, across G20 countries.

The team had three goals in mind: to determine which tools produced the most reliable and useful research, how their outputs and sources could be systematically fact-checked, and what an effective end-to-end workflow for AI-assisted research should look like.

The findings were a timely warning. AI-generated responses often appeared polished and well-structured even when the underlying references were inaccessible, incorrectly interpreted or potentially fabricated, reinforcing the importance of evaluating AI's reliability as a tool for investment and financial research. In response, the team developed a five-step framework: scope, collect, verify, synthesise and present, offering a practical blueprint for anyone using AI in high-stakes research or investment contexts.

An error occurred while preparing your download

Regulating AI in healthcare

Although GenAI and large language models (LLMs) are growing more capable, the rules governing the use of AI in healthcare have not kept pace. LLMs can now handle open-ended tasks and generate diverse responses across applications but existing medical AI regulations were largely designed for older, narrower systems with predictable outputs. The flexibility that makes LLMs powerful also makes them harder to govern as they can hallucinate, produce inconsistent results and generate recommendations that are difficult-to-explain or audit.

It was this gap that led Caitlyn Tan, an NUS Medicine graduate from the Class of 2026, to examine whether existing regulatory approaches are equipped to manage LLMs used for clinical decision support (CDS).

Her research examines a range of safeguards that could help make GenAI safer in clinical settings. These include red teaming, in which AI systems are deliberately pushed into difficult or harmful scenarios to expose weaknesses before deployment; guardrails, which constrain unsafe or inappropriate outputs and retrieval-augmented generation, which allows models to draw on trusted and up-to-date sources. Rather than advocating for rigid rules which could stifle innovation, she makes the case for forward-looking, adaptive regulations that evolve alongside the technology, while maintaining patient safety as the priority.

Applications to UROP+ REx, UREx and SPUR are closing on 28 August. Click on the links to apply today.