Rio Tinto

The Context

Rio Tinto is one of the world's largest mining companies. When I joined in June 2019, the analytics team was still in its early stages. There was a lot of data, decades of it, but it was trapped in legacy systems. The tooling was fragmented. The opportunity was clear: bring modern ML to operations that had been running on intuition and manual analysis.

What I Built

Industrial Optimization

The most impactful work was applying ML to physical operations. I worked on optimizing titanium smelting furnaces using machine learning, finding patterns in process data that could improve yield and reduce waste. I also worked on mine feed optimization for a large open-pit mine, where even small improvements in scheduling and feed composition translated to significant savings.

Predictive Maintenance

I built models to diagnose and predict equipment failure. In mining, unplanned downtime is extremely expensive. The goal was to catch problems early enough to schedule maintenance before something broke.

The Intelligent Mine

I contributed to a proposal to bring ML to Rio Tinto's first Intelligent Mine. This was a strategic initiative to demonstrate what a fully data-driven mine could look like, from extraction to processing to logistics.

Data Infrastructure

A lot of the work was not glamorous. Before I could build any models, I had to get the data into a usable state. I developed Spark ETL pipelines to ingest data from legacy systems into AWS, making decades of industrial data available for modern analytics. I also led an internal project to review and evaluate commercial data science platforms, and designed a shared Python library for the global analytics team.

Promotion & Team Leadership

I was promoted to Specialist Data Scientist in April 2020 and led a data science team on analytics projects for the largest iron ore producer in Canada.

What I Learned

Working at Rio Tinto taught me that the hardest part of ML in industry is not the algorithms. It is the data. Legacy systems, inconsistent labeling, missing values, and data that was never designed to be used for analytics. Before you can do anything interesting, you have to build the foundation.

It also taught me to speak the language of the business. Mining engineers do not care about your F1 score. They care about tons per hour, furnace yield, and unplanned downtime. Translating ML results into operational terms is a skill I use constantly.