Led the strategy, architecture, and development of Matchbox, an open-source data linkage platform combining Bayesian methods with graph theory. Implemented Splink for probabilistic record linkage. Built on FastAPI and PostgreSQL, deployed on AWS with Docker and Terraform.

Designed and maintained production ETL data pipelines using Airflow. Orchestrated DAGs to ingest, hash, and link multi-gigabyte datasets from sources including the Companies House API and HMRC trade data dumps, transforming raw data into unified entities for analysis.

Built tools to improve the developer experience for analysts. Created and maintained dwutils, a Python library used by around 50 analysts that abstracts data access and caching. Maintain the department’s first data science cookie-cutter to standardise project setup and CI/CD.

Led machine learning engineering initiatives, maintaining ML tooling and collaborating with infrastructure engineers to implement GitLab runners and inferential model APIs on AWS.

Deployed and configured aider and Claude Code on the department’s secure analytical environment during a hackathon — the first AI-assisted development setup of its kind in the department.

Regularly present on subjects from open source to type hinting to upskill the business. Mentor several colleagues.

Feature engineering for a LightGBM model predicting company export propensity, evaluating across key segments. Used cross-government datasets and developed a Streamlit dashboard with Docker and MLFlow.

Led projects including a Wikibase knowledge graph prototype and a reproducible analytical pipeline.

Mentored analysts in engineering best practices, encouraging code sharing, version control, and reproducibility.

Chaired a cross-government machine learning book club.

Key AI engineer productionising the Redbox product while seconded to a multi-disciplinary team at i.AI, developing a bespoke LLM-backed application for civil servants. Used LangChain, LangGraph, Elasticsearch, Redis, Django, PostgreSQL and Docker.

Led development of agentic RAG, including prompt engineering for tool-calling and a redesign of metadata extraction and the Elasticsearch query.

Developed an early evaluation framework using DeepEval and RAGAS. Productionised through unit tests.

HMRC Apprentice of the Year 2021.

Used XGBoost to predict product commodity codes from unstructured text data, building an end-to-end pipeline from extraction and validation through to an operational system. Resulted in a £3m investigation pipeline.

Redesigned a Power BI app for the compliance directorate SLT and led a Power BI community of expertise, regularly presenting new techniques.

Used SQL and R to enrich team projects — including ONS postcode lookups and a rebuilt Charity Commission database for mailing list cleaning.

Managed and directed a team delivering the national email programme under significant pressure and tight deadlines.

Delivered results at scale — fundraising raised a record £5m during the 2019 General Election campaign.

Managed relationships with high-value individual supporters and legacy prospects — influencing without hard authority.

Worked with the data team to design RFM-based audience segmentation for the mailing — an early intuition for what I’d later formalise as clustering and feature engineering.

Built SQL dashboards and carried out basic data analysis — data instincts predating any formal training.