Simple Transformer

Simple Transformer is a from-scratch implementation of a decoder-only transformer language model, written with PyTorch and trained on Jane Austen.

• Character tokenisation
• Multi-headed masked self-attention
• Pre-norm & dropout regularisation

source

PPO Sandbox

PPO Sandbox is a lightly optimised implementation of PPO using Pytorch and OpenAI Gym.

• Asynchronous vector environments
• GAE rewards and advantages
• Minibatch training
• Comprehensive stats logged and progress videos recorded

source

COVID Futures

COVID Futures provides forecasts of COVID case numbers in Australian states.

• Simple custom-built ML data pipeline
• ResNet architecture (extended from this paper) implemented with PyTorch
• Website via Flask / Tailwind / billboard.js

source

Australian Fungi ID

Fungi ID predicts species identification for Australian fungi.

• Dataset assembled from iNaturalist
• Image classification with resnet34 via FastAI library

source

---

Research Background

• Towards a Total Synthesis of Strychnine - organic chemistry honours thesis (H1, 2010)