Work

Experience

• Senior AI Engineer

  May 2026 – present

  Collaborit

  A hands-on engineering role covering both the development work and the cloud infrastructure needed to deploy machine learning solutions that scale and hold up in production.

  My main project so far is an end-to-end document ingestion pipeline for technical maintenance manuals for mining equipment. It processes the manuals and indexes them into a PostgreSQL vector database (pgvector), which powers semantic search and retrieval over them. Many of the manuals are scans, so I built a custom containerised OCR service that the pipeline calls only when a document needs it. Keeping text extraction separate means it can scale independently of the rest of the pipeline and be changed without touching it.

  I deployed the full solution on AWS. ECS and EC2 handle compute; RDS (PostgreSQL) and DynamoDB handle storage; API Gateway, Lambda and load balancing make up the service layer; and Amazon Cognito handles user authentication.

• Team Lead, No-Code AI Platform & AI Accelerator

  Aug 2021 – Apr 2026

  BMW IT Hub South Africa

  At the BMW IT Hub I led two teams. The first was the AI Accelerator, an in-house consulting team for business units across the BMW Group — IT, Sales, Logistics, Production and others. We worked with each unit to identify where machine learning was worth applying, built the model, and then industrialised it so it ran as part of the business rather than as a one-off analysis. Much of the job was the cross-functional part: bringing business stakeholders, engineers and data teams to agreement on what was being built and how it would be judged.

  The second was the No-Code AI Platform team, which ran DataRobot for the group. The platform let teams prototype, train and deploy models quickly, shortening the path from an idea to something in production and making it possible to test whether a use case was viable before committing engineering time to it.

  The use cases ranged widely. I built forecasts of vehicle sales in the South African and global markets. For the planning department I forecast option take rates over a three-month horizon — how many X3 buyers would choose 21” rims, leather upholstery, or a particular paint colour, for example — to improve supplier ordering decisions. For the EU market I developed a model that predicts how many days a vehicle takes to reach the dealer after leaving the plant, based on the delivery route (which rail lines and ports it uses), the production region and the destination region, with the data extracted and prepared from SAP Datasphere.

  Computer vision was a significant part of the work as well. After a hurricane struck the Charleston port where finished vehicles were waiting to ship, I worked with plant personnel in Spartanburg, USA, to train a model that detects rust on vehicles during final quality inspection. I also supported object detection use cases across several other departments.