🧪 Research
SLIME Lab | University of British Columbia, Canada
2025.04 – Present
- Supervisor: Professor Jian Zhu
- Mechanistic Interpretability in speech-capable LLMs.
- Develop structured ablation experiments across speech/text inputs.
- Apply activation-patching and neuron-level interventions to trace behaviors in speech-enabled LLMs
- Topics: Speech and LLM Interpretability, activation patching, TransformerLens, HPC.
Human-AI Interaction Lab | University of British Columbia, Canada
2024.10 – Present
- Supervisor: Professor Cristina Conati
- Designed transformer-based model architecture to predict user cognitive traits from raw eye-tracking data.
- Trained deep CNN-GRU models & optimized attention mechanisms, boosting accuracy by 12% over baseline.
- Designed full preprocessing pipeline for time-series gaze data.
- Topics: deep learning, GRU-CNN, Transformers, human-computer interaction, XAI.