My research focuses on developing robust and responsible AI systems that can operate reliably in real-world environments with imperfect data. I specialize in instance-dependent noisy label learning, where I've pioneered novel graphical model approaches and adaptive estimation techniques that significantly improve model performance when training labels are corrupted or unreliable. Through my work at TikTok, I advanced multimodal large language models (MLLMs) for Trust & Safety applications, achieving substantial AUC improvements by integrating diverse vision backbones with state-of-the-art language models.

Currently, as a Grant-Funded Research Fellow at the Responsible AI Research Centre, I'm developing privacy-preserving machine learning frameworks for Australian data sovereignty and AI safety protocols for critical infrastructure. My research bridges theoretical advances with practical applications, from contributing ML innovations to major film productions (Mad Max: Furiosa, Wolverine & Deadpool) to architecting scalable AI systems with multi-million dollar revenue impact. I maintain active collaborations with leading international institutions and contribute regularly to top-tier conferences including ECCV, WACV, and ICCV.