Hi, I'm Rafael Chong.
JC2 at Hwa Chong, CSIT Computing Scholar. First-author arXiv on LLM perceptual limits in CAPTCHA, computer vision pipeline for chemistry lab analysis with NTU (SSEF Finalist, 91.1% accuracy), and winner of the 2024 National Sentinel Challenge.
Global Top 10 at the Global Youth Entrepreneurship Competition (430 teams, 28 countries), champion of NES Connect 2025. Outside: Overall Captain of Hwa Chong Track & Field, 2x NSG C Division Champion in 100m and 4x100m relay.
Profit-maximising RL agents endogenously generate leverage bubbles and cascading margin calls from a stable equilibrium. Features a live interactive Streamlit dashboard.
Introduced ASCII art and noise-corrupted audio as CAPTCHA classes. No frontier model (GPT-4o, Gemini) solved a single ASCII instance. arXiv preprint; peer review in progress.
Submitted to the John Locke Institute Economics Competition. Argues enrichment through institutional power, not mere theft.
Do profit-maximising RL agents endogenously generate financial fragility? Agent-based simulation of Minsky's Financial Instability Hypothesis with leverage, margin calls, forced liquidation, and an interactive Streamlit dashboard.
Do human-specialised neural pathways make CAPTCHA LLM-proof? Introduced ASCII art and noise-corrupted audio CAPTCHAs - no frontier model (GPT-4o, Gemini) solved a single one.
Obstacle detection and avoidance on a DJI Tello drone using deep learning depth estimation and image processing, under Asst Prof Mir Feroskhan at NTU.
Investigated the viability of human-specialised neural processing as a basis for next-generation CAPTCHAs resilient to multimodal LLMs. Introduced two CAPTCHA classes (ASCII art-based visual CAPTCHAs and noise-corrupted audio QA CAPTCHAs) and evaluated them against frontier models including GPT-4o and Gemini. No model solved a single ASCII-based CAPTCHA and all models performed near-randomly on audio CAPTCHAs, suggesting strong near-term robustness against current AI capabilities.
LabAssist: a computer vision system for detecting procedural errors in chemistry titration experiments. Investigated synthetic video data to improve model generalisation, comparing low-fidelity (background replacement, GMFlow optical flow) and high-fidelity (diffusion-based image animation) pipelines. GMFlow-based augmentation achieved 91.1% accuracy vs. an 83.3% real-world-only baseline, demonstrating that motion-preserving synthetic data generation is an effective strategy for scalable training dataset expansion.
Vice President
Student (Economics & Management Stream)
Technical Director (Machine Learning)
Machine Learning Engineer
Research Intern
Integrated Programme (Junior College) · Promotional Exams: AAAA/B
Integrated Programme (Secondary) · MSG 1.30, 9 A1s, Higher Chinese O-Levels A2M
Awards, scholarships, and competition results.
AI Singapore
Highest award tier; placed among the top 12 students nationally at the pre-university level in AI concepts, applied ML, and computational problem-solving.
Science Centre Singapore
Presented original research on computer vision and synthetic data generation for chemistry lab error detection at Singapore's national science and engineering competition.
Equinor · Space Centre Houston
Recognised at the Conrad Challenge, a prestigious international STEM innovation and entrepreneurship competition supported by Equinor and Space Centre Houston.
Ministry of Education Singapore
Awarded to top-performing students in independent schools in recognition of outstanding academic achievement and conduct.
Centre for Strategic Infocomm Technologies
One of Singapore's most competitive pre-university computing scholarships, granted to students with exceptional aptitude in computing and technology.
Ministry of Education Singapore
Awarded by the Ministry of Education Singapore in recognition of strong academic performance and character at the pre-university level.
Entreplanet
Represented Singapore and placed Top 10 out of 430+ teams from 28 countries in the world's largest youth entrepreneurship competition.
NUS Entrepreneurship Society
1st place at one of Singapore's most prominent pre-university entrepreneurship competitions, pitching an original startup concept against teams from across the country.
Amazon Web Services · Delta Careers
1st place in the internal data analytics case study competition among all participants of the AWS Work Experience Programme.
Digital Intelligence Service, MINDEF Singapore
1st place nationally in Singapore's premier Capture-the-Flag cybersecurity competition, organised by the Digital Intelligence Service under MINDEF.