About Me
ML engineer focused on production retrieval systems, AI agents, and predictive modelling. Currently shipping ML features at CompuCal in Cork — BSc from MTU Cork just wrapped.
My programming started in 2021 with Python scripts for crypto trading, which got me interested in blockchain and, later, robotics on the side. Those are past curiosities now — production ML is where the work actually is.
My Journey
Education
BSc Software Development, MTU Cork — completed. QQI Level 6 Advanced Software Development with distinction prior. Focused throughout on systems, algorithms, and applied ML.
Current Role
Shipping ML features at CompuCal in Cork — a regulated calibration-management platform. Production RAG, OCR pipelines, predictive maintenance models, and MCP servers that give internal agents real database access.
Focus
Focused on production ML — retrieval systems, AI agents, and predictive modelling for tabular and document-heavy problems. Comfortable owning a feature from research spike to deployed endpoint.
Technical Arsenal
ML & AI
- • PyTorch
- • scikit-learn
- • CatBoost
- • XGBoost
- • LightGBM
- • pandas
- • NumPy
GenAI & NLP
- • LangChain
- • CrewAI
- • MCP
- • FinBERT
- • Transformers
- • OpenAI API
- • Gemini
Backend
- • Python
- • FastAPI
- • Node.js
- • TypeScript
- • Next.js
- • Golang
- • Rust
Cloud & Data
- • PostgreSQL
- • pgvector
- • Azure
- • AWS
- • Docker
- • Redis
- • GitHub Actions
Areas of Expertise
RAG & Retrieval
Embeddings, hybrid search, confidence scoring, and citation surfacing. pgvector in Postgres for most cases, LangChain for orchestration, FastAPI for serving. I care more about evaluation harnesses and retrieval quality than the latest framework.
AI Agents with Tool Access
Multi-agent orchestration via CrewAI. MCP servers that give agents real read/write access to SQL databases, internal APIs, and documentation. The hard part is the tools and evaluation — the prompts are the easy bit.
Predictive ML
Classification, regression, anomaly detection, and forecasting on tabular data. CatBoost, XGBoost, LightGBM, and ensembles. I benchmark against real baselines with proper validation splits, not just hold-out on a toy dataset.
Document Extraction & OCR
Turning scanned PDFs and messy documents into structured JSON. Azure Document Intelligence when accuracy matters, in-house GLM-OCR when cost matters, and a lot of careful post-processing either way.