Founder of ADL Mechanical Services Ltd · completing an MSc in Computer Science with AI (St Mary's University, April 2027) · open to AI/ML conversations, collaborations, and consulting.
Seventeen years running mechanical engineering work — the last few of them building ADL Mechanical Services Ltd — and now applying that same systematic, debug-under-pressure mindset to machine learning and data science.
The MSc is the formal route in. The day job is where the discipline came from.
- ML / data — Python, PyTorch, TensorFlow, scikit-learn, Hugging Face Transformers, pandas, NumPy
- NLP / CV — Transformers (DistilBERT, RoBERTa, ALBERT), MediaPipe, YOLOv8, OpenCV
- Web — Next.js, React, TypeScript, Tailwind CSS
- Backend — Node.js, Express, Supabase, PostgreSQL, SQL
- Other — Git, Docker, Homebrew, Arduino, Klipper, 3D printing
- Jarvis — voice-first local AI workstation for Apple Silicon (Whisper + Ollama + Kokoro, 76+ tools, MCP server). Daily driver, white-label, MIT.
- MSc final stretch — transformers, NLP, computer vision, end-to-end ML pipelines
- Interpretable and ethical model design
Jarvis — Voice-first AI workstation with an instrument-cluster HUD. Local Qwen + Whisper + Kokoro on Apple Silicon, 76+ voice-callable tools, hot-reloadable plugin system, white-label via config/brand.json. Docs at aoneill.co.uk/arc.
Claude-Office — Pixel-art isometric office that visualises AI agents collaborating live over WebSockets. React + TypeScript + Vite on the front end, Express + ws on the back, SQLite for state.
Twitter Sentiment Classification — Entity-level sentiment classifier across six architectures (MLP, BiLSTM, 1-D CNN, DistilBERT, RoBERTa, ALBERT). MLP topped the leaderboard at 98.6%, beating RoBERTa at ~1/70th the training cost. MSc NLP final assessment.
Chest X-ray Classifier — Pneumonia classification with knowledge distillation: ResNet-50 teacher → MobileNetV3 student. MSc Machine Learning final assessment.
COBOL ATM — Full ATM written in COBOL with a Flask bridge and a retro CRT interface.
Doom Scroll Detection — Webcam-based posture and attention monitor built on MediaPipe and YOLOv8.
Snake RL — Deep Q-Network agent with experience replay, learning to play Snake from scratch. MSc AI final assessment.
Heart Disease Classification — Random Forest, SVM, and neural network comparison reaching 85% accuracy.
NCM Classification Practical — Interactive notebook walkthrough of NCM, LDA, and QDA classifiers.
Conversations around AI/ML, Data Science, and applied machine learning — particularly anything bridging engineering domain knowledge with ML. Open to collaborations, consulting, research, and the right full-time AI/ML role from April 2027. Best way to reach me is LinkedIn.




