My research spans five interconnected areas at the intersection of Human-Computer Interaction, AI-supported learning, and interactive technologies.
This research investigates how AI can act as an interactive partner in learning, supporting reasoning, reflection, and skill development through human–AI collaboration. Key projects include:
A multi-agent LLM tutoring platform that addresses “vibe coding” — the tendency for students to rely on AI-generated solutions without understanding — by reframing AI from answer generator to Socratic coach. The system pairs a Driver Agent that delivers strictly non-directive feedback in four tiers (analysis → hints → questions → suggestions) with a Navigator Agent that injects plausibly flawed code calibrated to the student’s reasoning level, requiring them to evaluate, debug, and accept or reject the AI’s output. The design is grounded in Cognitive Load Theory (tiered scaffolding directs effort towards germane load) and Self-Regulated Learning (Driver mode supports metacognitive monitoring; Navigator mode supports critical evaluation).
Empirical evidence (NTU SC1007, Jan–Apr 2026): 1,000 active students, 10,333 sessions, 35,433 code submissions across 14 lab questions over 7 lab weeks. Mode-level differences are statistically distinct (Kruskal–Wallis H = 886.01, p < 10−192): Navigator Mode produces the most efficient trajectories (1.80 attempts to first accept) while Driver Mode produces the highest-quality submissions (5.13/10 tests passed). An accompanying SRL survey (N = 155, 12-item scale) returned an overall mean of 3.39/5 with all items significantly above the neutral midpoint — Reflective Self-Evaluation strongest (M = 3.46), Checking Understanding the largest individual effect (t = 6.97).
Reported in Beyond the Perfect Assistant: Provoking Learning with Flawed AI Partners (ACM DIS 2026) and at NIE RPIC 2026. Built on a stateless API architecture (OpenAI backend, React frontend, Azure deployment) so the pedagogical pattern is domain-general and replicable. Funded: NTU-EdeX S$40,000 (2025–2027) · IRB-2023-735 · algogpt.azurewebsites.net →
Multi-agent AI platform simulating realistic technical interview environments with live coding, speech interaction, emotion recognition, and behavioural analysis. Provides structured, personalised feedback on technical performance and communication skills. Technology Disclosure filed: NTU Ref 2025-208
Learning analytics system visualising learner behaviour, progress, and topic dependencies for AI chatbot-based learning. Features peer comparison, self-referenced progress tracking, and structured feedback mechanisms. (NIE RPIC 2026)
Multi-agent AI simulation enabling teaching assistants to interact with AI-driven student agents representing diverse learning behaviours, supporting development of facilitation and instructional skills at scale.
Generative AI chatbots designed to support graduate students’ critical reasoning and academic writing. Co-PI: NIE I3G Grant S$104,897 · (NIE RPIC 2026)
Application of LLMs and computer vision to automatically analyse classroom lessons for spatial pedagogy and lesson microgenres, producing pedagogical heatmaps to support educator reflection. (NIE RPIC 2026)
Exploring how interactive technologies enhance learning across diverse educational contexts, integrating technology with pedagogical principles to deliver accessible, personalised learning environments.
Designing and analysing human-centred interactive systems for educational contexts. Research investigates affective and behavioural signals during learning, multimodal interfaces for instruction, and child-focused interactive systems for play and learning.
Designing and analysing interactive, human-centred systems beyond the classroom — including interactive media, mobile applications, novel interaction techniques, and ubiquitous computing.
Multimodal interfaces that combine audio, visual, and behavioural signals to support collaborative interaction, spatial awareness, and group activity understanding.
| Role | Period | Project | Amount (S$) | Funder |
|---|---|---|---|---|
| PI | 2025–2027 | AlgoGPT: Multi-Agent AI for Data Structures Learning | 40,000 | NTU-EdeX |
| Co-PI | 2025–2026 | GenAI for Graduate Critical Reasoning & Writing | 104,897 | NIE I3G Grant |
| PI | 2016–2018 | Socially Mediated AR for Enhanced User Experience | 148,000 | Edge Lab (AIA) |
| PI | 2017–2019 | Twittener: Twitter Speech Synthesis with NLP | 50,000 | MOE AcRF Tier 1 |
| PI | 2016–2018 | Code Quality Assessment Tool (CQAT) | 18,000 | NTU-EdeX |
| Co-PI | 2022–2024 | Migrant Deities: Sacred Pantheon in Singapore | 89,000 | MOE AcRF Tier 1 |
| Co-PI | 2018–2020 | Hidden Shrines of Singapore | 70,000 | NHB HRG Grant |
| Co-PI | 2017–2019 | Productive Failure via Educational Games | 285,866 | MOE-TRF Grant |
| Co-PI | 2010–2012 | AR for Military Applications | 1,880,000 | MINDEF Singapore |
Beyond publications, my research has produced commercialisable technologies addressing real-world problems in distributed systems, public health, and consumer applications. To date: 1 patent filed (Petimo), 20 technology disclosures filed at NTU, and 3 technology disclosures licensed to industry and global health partners. Selected licensed technologies are featured below.
Transaction latency is a core limitation of leading blockchain networks, preventing real-time processing in production systems. We embedded blockchain features — immutability and distributed consensus — into Apache Cassandra, a fault-tolerant, highly available, high-throughput distributed database. The result is a hybrid platform that retains blockchain’s integrity guarantees while operating at database-class throughput, suitable for enterprise transaction workloads.
NTU Tech Disclosure 2018-225 · W. K. Ng, E. Bandara, M. H. S. Tharaka, O. N. N. Fernando · Licensed August 2018
A mobile participatory platform for dengue public health surveillance, communication, and citizen engagement, launched in Colombo, Sri Lanka. Deployed nationally through the DengueFreeChild app, the system enables citizens and schoolchildren to proactively report suspected dengue cases, allowing public health authorities to identify mosquito-breeding hotspots and alert parents in affected localities.
Public health impact: Received widespread national publicity in Sri Lanka across television, radio, and print press (March–April 2018), with documented contributions to earlier outbreak signal detection and stronger community engagement around childhood dengue prevention.
NTU Tech Disclosure TD/248/14 · M. O. Lwin, O. N. N. Fernando, S. Vijaykumar, V. Ratnayake, S. Foo · Licensed September 2017 · Watch coverage →
A location- and interest-aware event discovery platform that recommends nearby events to users based on shared interests and proximity, while allowing anyone to create an event and broadcast it to potential participants in real time. The technology underpins ongoing commercial software consultancy services delivered through Newtonis Technologies Pte Ltd.
NTU Tech Disclosure TD/305/17 · O. N. N. Fernando, N. Y. H. Lee, W. K. Ng · Licensed March 2018