Spearheaded the development of a generative AI-powered audit automation tool at BDO UK, revolutionizing audit processes across Audit, Cyber Security, and IT teams. Leveraged Retrieval-Augmented Generation (RAG) systems with custom large language models (LLMs) built using LangChain and LlamaIndex, fine-tuned for domain-specific audit tasks. Designed and deployed scalable MLOps pipelines on Azure, integrating MLflow and Weights & Biases for experiment tracking and model optimization, achieving a 9x acceleration in audit task completion.
Technologies: LangChain, LlamaIndex, Azure, MLflow, Weights & Biases, PyTorch, vLLM, GCP, AWS, Google Colab
Outcomes: Reduced audit reporting time by 90%, automated 75% of repetitive tasks, and established a scalable framework adopted across multiple client offerings. Led a team of 4 (1 AI engineer, 3 data analysts), delivering webinars and seminars as the SME in ML and generative AI.
Architected and led the development of a full-stack ML optimization engine at ShopriteX, South Africa’s largest retailer, to maximize margin, revenue, and profit. Built predictive models using PyTorch and TensorFlow, incorporating transformer architectures and deep neural networks, deployed via end-to-end MLOps pipelines on AWS, Azure, and GCP. Utilized JIRA for project management and PowerBI for executive-level visualizations, driving data monetization initiatives that transformed business decision-making.
Technologies: PyTorch, TensorFlow, Keras, JAX, AWS, Azure, GCP, JIRA, PowerBI, Git, Docker
Outcomes: Boosted group margin by 15% and revenue by 20% within 18 months, impacting a $20B+ retail operation. Led a team of 7 (4 data scientists, 3 data analysts), ensuring scalability and maintainability of solutions with CI/CD practices.
Developed a machine learning-based transaction monitoring tool for Glacier by Santam’s forensics team to detect and prevent fraudulent activities. Engineered predictive models using Scikit-learn and Python, deployed on SAP Hana for production-scale performance. Optimized data pipelines with SQL and SAP integrations, enabling real-time fraud detection across millions of transactions.
Technologies: Python, SQL, SAP Hana, Scikit-learn, PowerBI
Outcomes: Enhanced fraud detection accuracy by 30%, preventing losses exceeding $5M annually. Streamlined forensic workflows, reducing investigation time by 40%.
Designed an innovative ML optimization model at Glacier by Santam to strategically allocate assets between Life and Living annuities, enhancing client retirement outcomes. Integrated legacy mainframe data with external sources (MorningStar, Bloomberg) using Python, R, and Excel/VBA, applying advanced statistical techniques to reassess actuarial persistence assumptions and optimize annuity strategies.
Technologies: Python, R, Excel/VBA, SQL, MorningStar API, Bloomberg
Outcomes: Improved retirement benefits for over 10,000 clients, increasing satisfaction by 25%. Released $10M+ in actuarial reserves for reinvestment by aggregating client mortality data.
Constructed advanced Business Intelligence (BI) cubes at Glacier by Santam, modeling new dimensions and facts to deepen data insights. Leveraged Python, SQL, and SAP Hana to integrate disparate data sources, enabling comprehensive analytics on client retention, churn, and segmentation. Delivered actionable visualizations to support strategic decision-making for executive stakeholders.
Technologies: Python, SQL, SAP Hana, PowerBI, Excel
Outcomes: Increased data accessibility by 50%, reducing reporting latency by 60%. Empowered leadership with insights that drove a 15% improvement in client retention strategies.
Led the development of actuarial valuation and pricing models at Glacier by Santam as part of the pricing committee. Utilized R, Python, and SQL to construct sophisticated models assessing the impact of rate strategies on client portfolios, optimizing Life product offerings. Integrated market data from MorningStar and Bloomberg to inform pricing decisions, enhancing competitive positioning.
Technologies: R, Python, SQL, MorningStar, Bloomberg, Excel/VBA
Outcomes: Optimized pricing strategies, increasing product profitability by 12%. Influenced advisor strategies, improving client portfolio performance by 18%.
Conducted performance attribution analysis at Sanlam Investments, evaluating unit trusts across South African and global markets. Utilized Excel, VBA, RStudio, MS Access, MorningStar, and Bloomberg to perform a 60/40 split of qualitative and quantitative research, delivering detailed due diligence reports to inform investment strategies for board-level decisions.
Technologies: Excel, VBA, RStudio, MS Access, MorningStar, Bloomberg
Outcomes: Enhanced buy-list recommendations, improving portfolio returns by 10%. Streamlined due diligence processes, reducing analysis time by 25%.