Modernised the orchestration of analytical data pipelines by establishing a dedicated Apache Airflow environment for the Advanced Analytics team and migrating dbt execution from a legacy orchestration setup to Astronomer Cosmos.

Designed a repository and containerised execution environment that integrates dbt with Airflow and supports independent deployments, giving the team greater ownership while reducing reliance on shared infrastructure.

Led the evaluation and proof of concept for Cube as a centralised semantic layer, exploring reusable metrics, pre-aggregations, BI integrations and AI-assisted analytical querying. Authored an RFC documenting the architecture and trade-offs.

I am also actively involved in the rollout, helping establish a governed, consistent way for teams and analytical applications to access business data.

Developed analytical data models to support the evaluation and comparison of fraud risk scoring systems used within identity verification workflows.

Applied dimensional modelling to connect verification events, global, network and device-based fraud signals, and outcomes, enabling consistent comparisons of model effectiveness and overlapping detection mechanisms.

Developed and improved production LLM-powered services used to extract and structure information from construction industry data.

Integrated language models and structured outputs into existing workflows for email classification, project phase extraction and article titles and summaries, improving the consistency of extraction from unstructured information.

Developed a centralised proof-of-concept framework for evaluating the quality and reliability of production LLM services.

Using Langfuse, the framework combines LLM-as-judge techniques with deterministic and quantitative metrics. Separating evaluation logic from individual services provides a reusable way to assess performance, identify regressions and compare changes across applications.

To support the transition from a single-client setup to a multi-client model, I built a centralized Terraform repository for our analytics infrastructure on GCP. Working closely with Google’s DevOps team, I implemented CI/CD using Cloud Build to automate and standardize environment creation. This transformation allowed us to scale efficiently, spin up client-specific environments with ease, and significantly improved the reliability of our cloud infrastructure.

To address persistent data quality issues, I implemented the dq-utils dbt package to log failing test results and link them to the original source data. I then built a Looker dashboard that provided real-time visibility into data issues, trend analysis, row-level drilldowns, and Jira integration for issue tracking. This solution enabled proactive data ownership, alerting system owners when quality thresholds were breached, significantly improving the overall reliability of our data.

I designed and built a unified data model in dbt to consolidate all customer “trade” events, including sales, package changes, and churn, into a single source of truth. Collaborating with operational teams, I mapped user journeys to define event logic, implemented rigorous testing and documentation, and created a Looker dashboard that now serves as a trusted operational reporting tool. This solution enables the business to track customer movement, identify underperforming sales channels, and make data-driven decisions with confidence.

To address inconsistent and duplicated reporting requests, I developed a lightweight intake process based on BEAM data modeling principles. This involved collaborative sessions with stakeholders to define key user stories and metrics, an Excel mock dataset to map expected dimensions, and an ER model to assess feasibility. The process improved clarity, reduced duplicate requests, and enabled better prioritisation — ultimately streamlining reporting workflows and strengthening alignment between data and business teams.

I designed and developed a website for a jungle trekking company in Sumatra, aimed at boosting their online presence and increasing direct bookings. Over a 6-week sprint, I led the project end-to-end — from photography and design to content strategy and marketing — gathering regular feedback to iterate quickly. The result was a visually engaging, user-friendly site that significantly improved the company’s online booking rates and became a major driver of new income.

I’m building an end-to-end Fantasy Premier League analytics platform, ingesting data from the FPL API into BigQuery with dlt and Dagster, then transforming it with dbt.

A Cube semantic layer powers player and team analysis in an Astro dashboard, bringing data engineering, modelling and visualisation together.

Analytics engineering business reporting pipeline built for a large telecommunications company. Written using best-practice data modelling, robust testing, and comprehensive documentation. Features include incremental models to handle big data, separate environments for dev/prod, testing legitimacy of manually entered data as pre-hook and BQML analysis running at the end of the pipeline.