- Supervisor: Dr. Ashton Anderson- Simulated real human individuals with LLMs through fine-tuning and retrieval-augmented generation (RAG).

- Supervisor: Dr. William Cunningham- Led the development of Gem, a reinforcement learning (RL) framework designed specifically to facilitate researchinto psychologically plausible interactions between agents and environments, using PyTorch and NumPy.- Collaboratively developed Minagen, a LLM agent framework designed for building agents with human-like cognitivearchitectures.- Optimized numerical representations of agents and environmental objects to improve runtime of experiments by 200%.◦ Refactored existing codebase to align with standard coding practices, achieving a 50% reduction in code volume.

- Supervisor: Dr. Haohan Wang- Led a study as the first author focused on the enhancement of inherent adversarial robustness within condensed datasets through curvature regularization, leading to improvements in both clean accuracy and robustness- Designed a LLM-based AutoML system to automatically train and evaluate models for vision tasks, achieving 20% improvement in accuracy over code produced by vanilla LLMs.

- Worked on the development of the 2D object detection system on an autonomous vehicle.- Optimized YOLOv5 models to improve detection accuracy by 10% over previous performance.- Collected and labeled images for a comprehensive traffic-oriented image dataset.- Integrated Weights & Biases into the training pipeline, reducing the timeline for hyperparameter experiments by 75%.

- Worked on development of a knowledge graph-based gene annotation tool at a biotech startup in HK, China- Wrote a parser tool to convert JSON data to Atomese (a knowledge representation language in OpenCog Atomspace)