Projects

Developed the Local Adversarial Refinement Framework (LARF) to identify vulnerabilities in 16 local LLMs, including Lllama2, SmoLLM and Deepseek-V2, using models like GPT-4 and Llama-2 for attack and evaluation, improving safety in Generative AI.

Designed and implemented a pipeline integrating Google MediaPipe with game engine functionalities, achieving sub-2-second latency from input capture to display for real-time, AI-driven 3D facial motion capture across multiple subjects.

Researched and evaluated 2D and 3D facial motion capture techniques, identifying limitations and proposing solutions to enable real-time, accessible applications in public settings.