RMIT GenAI & Cyber Security Hackathon
A multi-phased adversarial machine learning competition tackling AI safety, prompt detection, and interactive simulation. Ranked 1st among 600+ participants.
1st Place Winner (Melbourne Campus) — outperformed 70+ teams across 3 global campuses in AI safety and red teaming challenges.
RMIT GenAI & Cyber Security Hackathon
Spearheaded an ensemble deep learning approach using DistilBERT, RoBERTa, and DeBERTa to detect unsafe prompts from a 5,000-sample dataset.
Achieved high validation AUCs through a hybrid TF-IDF and deep learning architecture, optimizing performance with early stopping and dynamic ensemble weighting.
Rigorously stress-tested Microsoft Azure OpenAI’s safety filters by successfully bypassing 2 of 5 ultra-advanced prompts, revealing key vulnerabilities.
Architected and developed 'Rising Waters' — a custom interactive web game that simulates a flood crisis using limited resources, blending strategy and tech.
Ensemble Deep Learning for unsafe prompt detection
Adversarial Red Teaming against Azure OpenAI filters
Interactive Crisis Simulation development
Hybrid TF-IDF & Deep Learning architecture