Cat Used Pricing Analytics

I worked with colleagues at Equal Experts on a commission from Caterpillar to design, build and maintain the Cat Used Pricing Analytics tool (CUPA) which provides price estimates for used equipment to Cat dealers.

I helped transform the tool from a prototype into a production system using a variety of MLOps techniques. I retrofitted 100s of tests to accelerate development and reduce the defect rate. I re-architected a complex network of interacting processes and API calls into a simple, linear batch-processing pipeline to ensure that results are reproducible across environments and that data provenance is traceable. This made the system more robustness and reduced the time lost investigating problems. It also enabled us to fork branches of the data for parallel development and experimentation letting the team scale horizontally without bottlenecks.

I provided a statistical methodology to support product development, translating user requirements into modelling requirements. I set out a roadmap to take us from black-box machine learning optimisation into a statistical model with analytical foundations that allowed us to test and explore new ideas and data sources.

I also wrote analytical reports to answer business queries in support the development of a new strategy for Cat’s Used and Rental business unit. I helped colleagues to understand the often counter-intuitive implications of statistics and sought to inspire others about the possibilities of modern scientific methods.