01 / the whole timeline

Work

Banking, IoT and retail. Rules engines, streaming platforms, RAG systems and computer vision — all of it in production, where it has to keep running on a Tuesday afternoon.

Download the PDF CV

Experience

8 roles

Machine Learning Engineer

Capitec Banknow

09/2023 — Present · Sandton, South Africa

capitecbank.co.zaThe leading bank in South Africa by number of clients and user satisfaction.

Built and open-sourced Decider, a rules engine productionised for business credit granting, retail credit granting and fraud detection across 3 teams, executing rules on 7M+ transactions per day; presented the project at PyCon.
Established the Kafka and Apache Flink patterns used for real-time stream processing of transactional data.
Developed a real-time person-detection (computer vision) proof of concept on branch CCTV cameras, analysing where customers spend time in selected branches to target reductions in counter waiting times.
Redefined the model deployment process around Docker containers, cutting deployment debugging from hours of hands-on support to minutes of self-service, and built CI/CD pipelines that automatically integrate jobs with the bank’s scheduling platform.
Drive hands-on enablement of multiple teams across the bank, supporting adoption of ML tooling and best practices.

AI Engineer (Part-time Consultant)

CohesionXnow

02/2024 — Present · Remote, South Africa

prod.vectormind.chatAn AI startup building customised RAG chatbot systems.

Develop a production RAG chatbot platform built on LLMs, embeddings and vector search, alongside my full-time role.
Migrated the platform from a single EC2 instance to Kubernetes, improving maintainability and reliability.
Rebuilt the RAG ingestion and training flow on a serverless architecture, improving retrieval accuracy, reducing hallucinations and eliminating always-on compute costs.
Implemented downloadable source references for chatbot answers, improving verifiability and user trust.

Product Engineering Manager (Machine Learning)

IoT.NxT (Vodacom/Vodafone)

03/2023 — 09/2023 · Centurion, South Africa

iotnxt.comAn industry-agnostic IoT platform with Gartner recognition.

Led and managed a team of 10 developers delivering AI and data analytics products for the IoT platform, owning planning, delivery and stakeholder communication.

Senior Machine Learning Engineer (Team Lead)

IoT.NxT (Vodacom/Vodafone)

06/2022 — 03/2023 · Centurion, South Africa

Expanded the data analytics product into a low-cost multi-tenant solution scalable to thousands of clients, directly enabling sales to 6 high-value clients.
Set up agile processes and CI/CD pipelines, reducing time from requirement to deployment by 75% (months to weeks) while improving product quality.
Served on the platform architecture design board and hosted monthly AI knowledge-sharing sessions across the team.

Machine Learning Engineer

IoT.NxT (Vodacom/Vodafone)

06/2021 — 06/2022 · Centurion, South Africa

Built a data analytics platform — lakehouse (Apache Iceberg), BI reporting and stream processing — storing and processing 1M+ records per minute in near real-time.
Developed a no-code platform enabling clients to easily create real-time ETL pipelines.
Secured the full stack, including OAuth2 integration, and remediated 352 CVEs.

Junior Machine Learning Engineer

IoT.NxT (Vodacom/Vodafone)

04/2020 — 06/2021 · Centurion, South Africa

Deployed real-time ML predicting energy usage of air-conditioning units in cellular base stations, verifying control logic that reduced energy consumption by 8%.
Built models to select optimal locations for alternative energy installations and to detect anomalous base-station telemetry, surfacing underperforming equipment.
Created APIs and custom UIs serving model outputs, turning predictions into actionable insights.

Data Scientist

Cognizance (MAC Mobile)

02/2019 — 03/2020 · Pretoria, South Africa

cognizance-vision.comA retail analytics platform delivering real-time, actionable insights across the retail chain.

Trained a computer-vision model detecting front-facing shelf products for planogram compliance, achieving 97% accuracy.
Optimised the model for serverless deployment, reducing detection latency from 15 minutes to 5 seconds.
Built web tools for efficient collection and labelling of voice and image data, cutting labelling time per sample by 90%.

Full-Stack Developer

MathU (Instasense/Uvirco)

01/2018 — 01/2019 · Pretoria, South Africa

Sole developer at an AI-driven math-education startup: built an offline-capable mobile app for remote areas with unstable connectivity, plus a rewards-based data-collection programme that gathered thousands of sample math problems.