About Me

I'm currently working as an NLP Engineer @ aveni.

I have a PhD from Heriot-Watt University, where my research was on applying natural language processing-based machine learning algorithms to programming languages.

Prior to that, I obtained a First-Class Master's degree (MEng) in Digital Electronics (now known as Electronics and Computer Engineering) from the University of Sheffield.

In my spare time I enjoy: reading, listening to podcasts, playing board games, walking my dog, and watching sports.

Contact

You can contact me on twitter or via e-mail.

TILs

Things I've learned, formatted as Python notebooks.

2024-12-15	Making LLMs smarter with structured generation and Outlines
2024-02-17	Introduction to information theory (from a machine learning perspective)
2024-02-10	Decision trees and random forests from scratch
2023-10-03	Bootstrapping confidence intervals
2023-09-27	Parallelism in Python
2023-09-07	Some practical use cases for the walrus operator
2023-08-28	Retrieval augmented generation
2023-08-27	Implementing TF-IDF from scratch
2023-08-06	A/B testing with the chi-squared test
2023-08-03	Fine-tuning a sequence classification model with LoRA and the peft library
2023-07-31	How to use Open AI's "function calling" ability
2023-07-30	What's the best way to strip punctuation from a string?
2023-07-29	Why is the softmax function "off by one"?

Projects

• clip-search: Text-to-image search with OpenCLIP, Docker, Flask, Faiss, etc. and a basic front-end.
• lexisearch: Retrieval augmented generation on transcriptions from the Lex Fridman Podcast.
• GloVedle: A version of Wordle using GloVe embeddings.
• Wordle Terminal: A Python port of Wordle which is played in the terminal.
• UUIDv4 Generator: Generate UUIDv4s in the browser, because I didn't like the way the top result on Google generated them.
• easynlp: A library for performing natural language processing - such as zero-shot classification, translation, named entity recognition, summarization, and question answering - inference on given data utilizing the pre-trained models from transformers.
• Sentiment Analysis Tutorial: A tutorial on how to implement some common deep learning based sentiment analysis (text classification) models in PyTorch with torchtext, specifically the NBOW, GRU, bi-LSTM, CNN and Transformer models. Somehow got popular and has quite a few stars.
• Sequence-to-Sequence Learning Tutorial: A tutorial implementing neural (deep learning based) sequence-to-sequence models in PyTorch with torchtext, by implementing six NMT papers. Also has quite a few stars and was used as a basis for the official PyTorch language translation tutorial.
• Image Classification Tutorial: A tutorial covering how to implement some deep learning computer vision models in PyTorch with torchvision. Covers: a basic multi-layer perceptron, LeNet, AlexNet, VGG and ResNet.
• A Tour of Optimizers: A tutorial on common optimization algorithms used for neural networks, including: SGD, Adagrad, Adadelta, RMSprop and Adam.
• Recurrent Attention Model: A PyTorch implementation of the Recurrent Attention Model from the Recurrent Models of Visual Attention paper.
• Glyphs of Dialogue: A small project combining ideas from GlyphNet and Dimensions of Dialogue.
• countworld: Generating synthetic datasets that deal with counting.
• numberworld: A reinforcement learning toy environment for task-oriented language grounding.
• snake: Snake in Javascript.
• difference: A timed mental maths game in Javascript.