Data Scientist · AI Builder · Ireland

Michael Brett

I build agentic AI systems that turn complex data into decisions — regulatory RAG agents, natural language dashboards, and automated data pipelines. A few examples below.

Scroll to explore

Data in. Decisions out.
Everything in between, engineered.

I build end-to-end AI systems — from raw data pipelines through to dashboards and agentic natural language interfaces. The output is always working software: something people can use to make better decisions faster, not a prototype that needs six more months of work.

The core of what I do is turning operational data into decision-support tools: RAG systems grounded in real document corpora, multi-agent workflows that automate structured reasoning, and dashboards that let people ask questions in plain English and get answers with charts. I've applied the same engineering approach across pharma, property, finance, and logistics.

I also have a strong data science foundation — three years of contract work in sales forecasting, time series modelling, predictive analytics, and dashboard reporting before moving into AI engineering full-time. That background means I'm comfortable with both the analytical and the engineering side of a problem.

Before data science, I spent two decades translating complex technical concepts for non-technical audiences — a skill that now shapes how I scope projects, communicate with stakeholders, and build tools that people actually use. I bring the same clarity to a client boardroom as I do to a codebase.

Tools & Technologies

From problem to production.
Independently.

🔍

Scope & Design

Understand the business problem, identify what data exists, and define what "done" looks like — before writing a line of code.

🔧

Build & Ship

End-to-end delivery: data pipelines, AI agents, APIs, and interfaces. Working software, not slide decks or prototypes that need six more months.

💬

Communicate & Iterate

Two decades of translating technical complexity into clear language. I speak to engineers, stakeholders, and boardrooms with equal fluency.

Live demo applications

Working systems I've built, all publicly accessible. They reflect the kind of AI engineering I focus on — specific, grounded, and practically useful.

01 / Property Analytics

Irish Property Price Register Dashboard

Live market data · AI agent · AutoML forecasting

Live Demo

A live data pipeline ingesting and transforming transaction records from the national Property Price Register into 8 KPI cards, interactive charts, and an AutoML forecasting model — updated in real time. On top of that sits an AI agent layer: ask questions in plain English and get back generated Plotly charts or written market analysis, powered by Claude Opus 4.6. Fully accessible via the link above — no API key required.

8 live KPI cards updated from propertypriceregister.ie
7 interactive tabs with 3D charts and animated choropleth maps
AutoML forecasting with full model leaderboard
AI agent mode: ask questions in plain English
Agent returns generated Plotly charts or written analysis
Powered by Claude Opus 4.6 via LangChain tool use

Architecture

PPR Website Data Pipeline AutoML Models LangChain Agent Plotly / NL Output
Python Streamlit Plotly LangChain Claude Opus 4.6 statsforecast scikit-learn

02 / Pharma Regulatory

Pharma Regulatory Guidance Agent

RAG · Document grounding · Compliance benchmarking

Live Demo

A RAG system grounded in ~60 ICH, FDA, EU GMP, and EMA regulatory documents. Designed for scientists and regulatory affairs teams who need precise, citable answers from a large document corpus — not confident hallucinations. Every answer carries a confidence score and exact section citations. Fully accessible via the link above — no API key required.

~60 ICH / FDA / EU GMP / EMA documents in ChromaDB vector store
Confidence-scored answers with exact section citations
Dataset Benchmarker: upload analytical data for regulatory assessment
Scored compliance checklist with priority recommendations
Guideline quotes surfaced inline with each finding
Model selector: Claude Sonnet / Opus / Haiku

Architecture

60+ Regulatory Docs ChromaDB Vectorstore RAG Retrieval Claude LLM Cited Answers + Scores
Python Streamlit LangChain ChromaDB Claude API RAG

03 / Energy Forecasting

Solar Production & Energy Forecasting Agent

Time series ML · Weather integration · Agentic monitoring

Coming Soon

An end-to-end ML forecasting pipeline using real solar panel production and household energy consumption data. Multiple models (Prophet, XGBoost, LightGBM) compared via walk-forward validation with exogenous weather features from Open-Meteo. An agentic monitoring layer detects model drift, flags anomalies, and provides natural language energy management recommendations.

Real solar production & consumption data at daily granularity
Exogenous weather features via Open-Meteo API
Multi-model comparison with walk-forward validation
Agentic monitoring: drift detection & anomaly alerts
NL interface: "What's my expected production next week?"
Full ML lifecycle: ingest → train → evaluate → serve → monitor

Architecture

Solar & Grid Data Weather API Feature Pipeline ML Models Agent Monitor
Python XGBoost Prophet LangGraph Open-Meteo API FastAPI Docker

Interested in the work?

I'm always happy to talk through how these systems work under the hood — architecture decisions, RAG strategies, or agentic design patterns. If you're building something similar or want to collaborate, drop me a line.

micl.brett@gmail.com

Based in Ireland