CV
Profile
Applied machine learning researcher and AI leader specialising in speech and language systems, with a focus on continual and meta-learning for acoustic modelling. My work addresses catastrophic forgetting in few-shot speech classification, analyses representation geometry to understand and predict downstream task performance, and develops principled evaluation frameworks for real-world speech systems. I bridge theoretical research and production-grade implementation by designing reproducible experimentation pipelines, multilingual benchmarking strategies, and signal-quality diagnostics such as SNR-based filtering. In addition to research, I lead scientific strategy and ML architecture development in an industrial context — guiding research direction, defining experimental standards, mentoring researchers, and translating algorithmic advances into scalable deployed systems.
Core research themes
- Meta-learning
- Few-shot adaptation, MAML-family optimisation, and learning-to-learn frameworks for data-efficient systems.
- Continual learning
- Stability–plasticity trade-offs, catastrophic forgetting mitigation, and sequential task adaptation.
- Metric & representation learning
- Embedding geometry, manifold diagnostics, and principled evaluation of learned representations.
Experience
ByteFuse is an AI R&D company backed by a R55M investment from Novus Holdings, building enterprise-scale AI solutions across education, speech, and analytics.
- Drive overall AI and scientific strategy for the company: set research direction, define experimental methodology, and establish quality standards across all ML workstreams.
- Lead and mentor a team of researchers and engineers, bridging the gap between academic rigour and production delivery.
- Architect multilingual speech and language evaluation frameworks spanning 30+ languages, including SNR-based dataset filtering and benchmarking pipelines.
- Co-developed Maski, an AI-powered WhatsApp-based tutoring companion built with Maskew Miller Learning, serving 100,000+ South African learners with CAPS-aligned, personalised learning support.
- Own end-to-end ML architecture: from research prototyping through to production deployment with real-time multi-model orchestration and mobile-first constraints.
Researched and applied emerging ML and AI trends, particularly in deep learning, while contributing to the incubation of what would become ByteFuse AI.
Founded and led a startup providing comprehensive digital solutions for pet owners. Venture closed due to the COVID-19 pandemic.
Applied data science and machine learning to enhance customer experience across core network performance systems.
Education
- MSc, Data Science / Statistics — University of Cape Town2018 — 2020
- BEng, Electronic Engineering — Stellenbosch University2014 — 2017
Technical expertise
- Machine learning:
- Continual learning, meta-learning (MAML, OML), few-shot learning, metric learning, representation learning, self-supervised pretraining.
- Speech & language:
- Spoken word classification, language identification, acoustic modelling, multilingual speech systems, contextual grounding.
- Evaluation & diagnostics:
- Manifold analysis (RMQM), intrinsic dimensionality, robustness under distribution shift, SNR-based data filtering, benchmarking pipeline design.
- Frameworks & tools:
- PyTorch, PyTorch Lightning, Python, experiment tracking, production ML pipelines.
Community engagement
- Speaker & Practical Instructor, Deep Learning Indaba 2022 — Africa's largest gathering of AI practitioners.
- Presenter, Deep Learning IndabaX South Africa — representing ByteFuse AI.
- Media: Featured on CapeTalk discussing Maski and AI in South African education.
- Open source: published code for MAMLCon (Interspeech 2023) and RMQM (ICML 2022 Workshop) on GitHub.