RMIT GenAI & Cyber Security Hackathon
A multi-phased adversarial machine learning competition tackling AI safety, prompt detection, and interactive simulation. Melbourne Campus Winner at RMIT CyberHack 2025.
Melbourne Campus Winner for AI safety and red teaming challenges at RMIT CyberHack 2025.
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