About me
My route into AI was not a straight line, and that is the point. I started as a mechanical engineer, spent nine years as a data analyst building failure-prediction models and reporting pipelines for industrial engineering teams, then completed an MSc in Applied Data Science at the University of Sunderland.
That background shapes what I build. I have watched real equipment fail, so my MLOps project predicts machine failure from sensor data. I have worked in compliance-heavy environments, so my RAG system answers questions over regulatory documents with citations, because an unsourced answer is useless in a regulated setting.
Right now I am deepening the production side of the craft through two intensive programmes: a Full Stack Generative AI and Agentic AI Bootcamp and a DevOps, MLOps and LLMOps Engineering Bootcamp. Everything I build is public: the code, the evaluation results and the write-ups.
Projects
Five shipped projects across two tracks: production AI systems (MLOps, evaluated RAG, agents) and business analytics (SQL, Power BI, cost modelling). Every repo runs, is tested in CI, and reports real numbers. Each one is a case study, not a tutorial.
Predictive Maintenance MLOps Platform
ShippedMachine failure prediction served as a monitored, containerised API: PR-AUC 0.888, a cost-tuned threshold catching 96% of failures, SHAP explanations on every prediction, and a from-scratch drift monitor. CI retrains the model on every push.
RAG Compliance & Incident Intelligence Assistant
ShippedGrounded question answering over UK health and safety guidance: a citation on every statement or a refusal, hybrid retrieval with rank fusion, and a golden-set evaluation harness that gates CI. Hybrid retrieval hits 100% of sources on the eval set.
Multi-Agent Security Operations Triage
ShippedLangGraph agents that classify, enrich and prioritise incident reports for human approval, with the production concerns most agent demos skip: schema guardrails with repair, PII redaction, cost budgets and trajectory evaluation (100% priority accuracy).
UK Property Deal Analyser
Shipped2.9 million Land Registry transactions analysed in SQL, joined to ONS rents, and scored with a transparent deal model that flags the districts worth an investor's attention. Ships a Power BI deal-finder dashboard: star schema, DAX measures, page spec.
Manufacturing Downtime & Maintenance Cost Analysis
ShippedWhich machines fail, what downtime costs, and where the maintenance budget should go: failure-mode KPIs and a documented cost model showing two failure modes carry 70% of a £2.3M cost. Excel KPI report plus Power BI dashboard, from one command.
Skills
Languages
- Python (advanced)
- SQL
- Bash
- TypeScript (basic)
Machine Learning
- PyTorch, scikit-learn, XGBoost
- pandas, NumPy
- Feature engineering
- Model evaluation
GenAI & LLMs
- RAG and vector databases
- LangChain, LangGraph
- Prompt engineering, fine-tuning
- LLM evaluation (RAGAS)
MLOps & Backend
- FastAPI, PostgreSQL, Redis
- Docker, GitHub Actions
- MLflow, drift monitoring
- AWS, Linux
Analytics & BI
- Power BI, DAX
- SQL (DuckDB, window functions)
- Excel reporting
- KPI design, cost modelling
Writing
Technical write-ups from each project: the decisions, the trade-offs and the numbers.
// first post shipping soon
The first write-up covers building the predictive maintenance pipeline: baseline model, MLflow tracking and what the drift monitor caught. Follow along on LinkedIn or watch the GitHub repos.