01
SIGGRAPH Asia 2020 · Art Gallery
We Can’t See the Rainbow in the White
A durational light installation: a projected white line is decomposed through a prism and a servo-driven system — white into rainbow, rainbow into magenta, magenta into a single, irreducible blue. Built with Arduino, prism and moving image, it treats light as a medium to be taken apart into controllable primitives, making the invisible visible. The instinct behind it — decomposing a visual phenomenon into elements a system can control — is the same instinct my research direction brings to generative tools.
Installation view — durational light work, 8min45s
02
Published · IEEE MetaCom 2025
MetaClues: Constructivist Learning via Progressive Clues from Generative AI in a Metaverse Environment
A generative-AI learning environment in a metaverse campus: image-to-text, reasoning-LLM fusion, text-to-image diffusion and image-to-3D, orchestrated as progressive clues rather than answers. Its multi-faceted relevance score deliberately rewards imaginative divergence as much as faithful replication — the AI provokes iteration instead of prescribing solutions. Pilot study with 19 participants.
In-world Arts & Technology tutor
03
MSc Innovative Lab Course final project
A Modular Web UI Grounding System for Automated Social Media Post Scheduling
A three-stage vision–language pipeline — retrained YOLOv8 element proposals, fine-tuned CLIP ViT-L/14 retrieval, and a Qwen2.5-VL-3B LoRA reranker — evaluated end-to-end and LoRA-adapted out-of-domain to a production site. Evidence that I can fine-tune, adapt and rigorously evaluate vision–language models.
hit@1 · OOD LoRA
14.8% → 60.7%
04
Under review · IEEE TALE 2026
MobiClues: Constructivist Learning through Gamified Discovery and Multi-Agent AI Tutoring
The mobile successor to MetaClues: gamified discovery coupled with a multi-agent RAG tutoring system — agents diagnose prior knowledge, scaffold within the learner’s zone, gate retrieval, and prompt reflection. The same design commitment, provoke without prescribing, carried into a multi-agent architecture on a device every student carries.
