Research methodology
Knowledge Distillation
A research platform for studying how response-priming prompts change knowledge distillation from a large language model teacher to a smaller local model.
- Python
- PyTorch
- LoRA
- Hugging Face
- GSM8K
I work on efficient machine learning, with a focus on knowledge distillation, inference systems, and optimization.
Computer Science + Combinatorics & Optimization at Waterloo
01
Research methodology
A research platform for studying how response-priming prompts change knowledge distillation from a large language model teacher to a smaller local model.
02
Inference systems
An experimental C++ inference scheduler built on llama.cpp for comparing FIFO, continuous-batching, and heuristic AIMD policies on Apple Silicon.
Learning systems
A convolutional neural network and reverse-mode autodifferentiation engine implemented in C++ without machine-learning libraries.
Machine-learning fundamentals
A modular neural-network framework written in C++ without machine-learning libraries, organized around networks, layers, and neurons.
03
CIBC Global Asset Management
Built evaluation and GPU-accelerated modelling infrastructure for comparing interpretable financial models, including memory-bounded PyTorch tooling for large datasets.
WAT.ai, Design Team
Leads an ML research and engineering team developing a world-model control stack for bimanual cloth folding.
Bachelor of Mathematics, Computer Science