About

South African-born actuary and data scientist with experience across insurance, finance, and retail. I build AI solutions by understanding what the technology can do, what the data actually shows, and how decisions get made in practice.

My career has taken me from South Africa to Australia, then to Boston for graduate study, and back to Australia. Working across different countries and business contexts has shaped how I approach problems and what it takes to bridge technical capability with practical execution.

I’ve led teams at Auto & General and Fidelity Investments, contributed open-source tools used in production systems, and published research on bandits and recommender systems. Fellow of both the Actuarial Society of South Africa and the Actuaries Institute of Australia. Master’s degree in Computational Science and Engineering from Harvard.

Experience

Head of Product AI, Auto & General

Leading AI initiatives across product, underwriting, and pricing to help provide affordable insurance to Australians. Building systems that enhance decision-making while operating within regulatory frameworks where explainability is fundamental. The work balances innovation with the rigorous standards insurance requires.

Director of Data Science, Fidelity Investments

Led a team responsible for personalization, recommender systems, and marketing optimization across digital and distribution channels. Approached personalization as a learning problem using contextual bandits to balance user engagement with preference discovery.

Data Science Consultant, Quantium

Started as a graduate analyst and developed foundational skills across insurance, banking, and retail sectors. Learned to translate between technical capabilities and business needs, building models that stakeholders understood well enough to trust.

Education

Master of Engineering, Harvard University

Computational Science and Engineering at the School of Engineering and Applied Sciences. Explored decision theory, optimization, and computing systems. Research involved numerical methods and large-scale parallel computing for studying optimal heat transfer in Rayleigh-Bénard convection.

Bachelor of Science, University of Pretoria

Actuarial Science and Financial Mathematics. Graduated with distinction. Rigorous foundation in mathematics and statistics with focus on developing realistic solutions to complex problems.

Actuarial Qualifications

Fellow of the Actuarial Society of South Africa and the Actuaries Institute of Australia. Completed Fellowship exams while working full-time.

Projects

Selective

White-box feature selection library supporting supervised and unsupervised methods. Addresses item selection for exploration in recommender systems through multi-objective optimization, balancing diversity and label coverage. Used in production environments where understanding feature importance matters.

MabWiser

Research library for rapid prototyping of multi-armed bandit algorithms. Supports context-free, parametric and non-parametric contextual models with built-in parallelization. Designed to handle real-world deployment challenges: changing contexts, delayed feedback, and explainable decisions at scale.

Mab2Rec

Recommender systems library integrating bandit-based algorithms for exploration-exploitation trade-offs. Makes learning objectives explicit and tunable, allowing teams to balance confident recommendations with discovering user preferences. Modular design integrates with existing recommendation infrastructure.

Publications

  • Optimized item selection to boost exploration for recommender systems
  • S Kadıoğlu, B Kleynhans, X Wang
    Integration of Constraint Programming, AI, and Operations Research, CPAIOR 2021
  • Active Learning Meets Optimized Item Selection
  • B Kleynhans, X Wang, S Kadıoğlu
    arXiv preprint arXiv:2112.03105
  • MABWiser: a parallelizable contextual multi-armed bandit library for Python
  • E Strong, B Kleynhans, S Kadioğlu
    2019 IEEE 31st International Conference on Tools with Artificial Intelligence (ICTAI)
  • MABWISER: Parallelizable Contextual Multi-armed Bandits
  • E Strong, B Kleynhans, S Kadıoğlu
    International Journal on Artificial Intelligence Tools 30 (04), 2150021
  • Automated predictive product recommendations using reinforcement learning
  • A Jain, D Gupta, S Shekhar, B Kleynhans, S Kadioglu, A Arias-Vargas
    US Patent 10,936,961

    Perspective

    The best AI solutions start with understanding how decisions actually get made, not how we think they should be made. Technical excellence matters less than whether a model fits how the business operates and inspires enough confidence to be used. In regulated industries especially, this means spending as much time on business domain, regulatory constraints, and organizational decision processes as on algorithms.

    Given how rapidly AI capabilities are evolving, adaptability matters as much as expertise. The technology we use to solve problems is changing, and the problems themselves are shifting as new capabilities emerge. I look for people who stay curious about these changes and can thoughtfully evaluate when new approaches actually serve business needs versus when established methods remain the right choice.

    Service

    Program Committee Member, IAAI-25 & IAAI-26

    Evaluating submissions for the Innovative Applications of Artificial Intelligence Conference focused on deployed AI applications and real-world impact.

    Volunteer, Actuaries Institute Australia

    Contributing as examiner for professional qualification exams and supporting development of the next generation of actuaries.

    Interests

    Outside work, I spend weekends with my wife and son exploring Brisbane, finding good pastries, ice cream spots, and friendly dogs. Rugby fan with a particular soft spot for the Springboks. I also climb regularly, drawn to the problem-solving aspect as much as the physical challenge. Both rugby and climbing are things I’m looking forward to sharing with my son when he’s ready.