Skip to main content

Antony O'Neill

Building with AI, ML, and data. I ship production web apps.

MSc Computer Science & AI student. Before that, 17 years as a mechanical engineer. I'm used to taking apart complicated systems to figure out what's actually broken. These days I'm doing the same with code and data.

Available for projects
Portfolio
Visitor
Antony O'Neill
Antony O'Neill
Full-stack Developer
Next.js · React · TypeScript · AI/ML
aoneill.co.uk
ID: AO-2025
About Me
Antony O'Neill

I'm an MSc Computer Science & AI student at St Mary's University. Most of my time goes into machine learning, full-stack development, and building AI systems that actually do something useful.

Currently digging into reinforcement learning and ethical AI. The mechanical engineering background helps more than I expected. Software bugs and broken machines turn out to be surprisingly similar once you start pulling them apart.

MSc CS & AI · St Mary's

Timeline

How I got here and what I've been doing.

Skills

Technologies and tools I work with.

ML & AI
Python
PyTorch
TensorFlow
scikit-learn
pandas
NumPy
Deep Learning
NLP
Transformers
Computer Vision
Reinforcement Learning
Web Dev
TypeScript
React
Next.js
Node.js
Tailwind
Supabase
PostgreSQL
Tools
Git
Docker
GitHub Actions
Linux
CI/CD
VS Code
Jupyter

Featured Projects

Things I've built and put in front of real users.

Agent Office logo

Agent Office

Pixel-art virtual office that visualises your AI agents working in real time. Watch Claude Code agents spawn, sit at desks, take coffee breaks, and chat in a Slack-style panel — isometric, animated, and popular on social media.

TypeScriptReactVite
Jarvis logo

Jarvis

Personal AI assistant CLI built around Claude. A bridge architecture lets it talk to your apps, run plugins, and keep conversational state across sessions. Docs are hosted at /arc so you can read it without cloning anything.

Node.jsTypeScriptClaude API
Deep Q-Learning Snake Agent logo

Deep Q-Learning Snake Agent

Reinforcement learning agent that learns to play Snake with a deep neural network. I ran a stack of experiments to see how network size, memory buffer, and environment changes actually affected learning.

PythonPyTorchDeep Q-Learning
Twitter Sentiment Classification logo

Twitter Sentiment Classification

Entity-level sentiment classifier across six architectures, from a plain MLP up to fine-tuned transformers. Twelve runs, head-to-head: a tuned MLP topped the leaderboard at 98.6%, beating RoBERTa at a fraction of the training cost.

PythonPyTorchHugging Face
Heart Disease Classification logo

Heart Disease Classification

Binary classifier predicting heart disease risk. I compared Random Forest, SVM, and Neural Networks across nine configurations. Random Forest came out on top at 85%.

Pythonscikit-learnTensorFlow
Keep It What logo

Keep It What

Full-stack e-commerce platform for personalised gifts. Real-time product customiser, Stripe checkout, and an admin dashboard for orders and inventory.

Next.js 15React 19TypeScript
Doom Scroll Detection logo

Doom Scroll Detection

Webcam tool that watches your posture and phone usage to catch the doom scroll. Uses MediaPipe for face and hand tracking, YOLOv8 for phone detection.

PythonOpenCVMediaPipe
View all projects →

Open Source

What I've been pushing lately.

Let's work together

I'm finishing my MSc in AI and looking at data science, ML engineering, and software roles. Happy to talk about graduate schemes, contracts, or anything interesting.

Get in touch