Experience

• 

  University of pretoria

  Aug 2011 - Aug 2012

  Residence management committee member

  Block Manager, Head of Social Committee & Webmaster.

• 

  Prime meridian solutions

  Jan 2012 - Dec 2014

  Founder

  This was my own business venture - services offered included a range of graphic design options (logos, posters, flyers, etc), the creation of technical diagrams, professional photographic (& retouching) services, a specialisation in film photography and the digitisation and restoration of 35mm film strips and slides. Map creation and other such GIS services were offered on a consulting basis as well.

• 

  Africa remediation technologies

  Aug 2012 - Dec 2014

  Junior gis consultant

  Compiling maps and technical diagrams for project proposals and tenders for the likes of Sasol, Fraser Alexander, LIBSA and the Department of Water Affairs.

• 

  Perdeby

  Mar 2013 - Mar 2014

  Photographer

  I generated news, feature, entertainment and sport visuals for Perdeby; South Africa's largest campus newspaper with a weekly readership of 10 000 printed copies & more than 30 000 online readers during my time there.

• 

  Eskom

  Jul 2013 - Feb 2015

  Junior gis specialist (bursar)

  Duties included (but were not limited to) creating a model of Eskom’s Low Voltage Network, developing and publishing web mapping services, completing quality assessments of commissioned remotely sensed images, validating metadata, creating training resources and presenting these courses; statistical analysis for the Integrated Annual Report; analysis reports comparing cadastre data purchased from different data providers & compiling usable census data shapefiles based on figures compiled by StatsSA. Show less

• 

  Esri south africa

  Mar 2015 - Aug 2017

  Gis specialist

  November 2015 - August 2017 - Patterns and PracticesI worked closely with internal teams to improve how we positioned, sold, implemented & supported Esri technology and solutions. I also worked strategically with key clients to help uncover the business value and implementation roadmap of Esri technology and solutions within their organisations. Functions included but were not limited to research and development following key technology trends, creating value-driven prototypes, driving knowledge and growth across Enterprise Solution components, driving virtual reality solution architecture practices, writing white papers, proposal writing, creating re-usable organisation templates, overseeing skills development of account managers and providing cost estimates for professional services.June 2015 - October 2015 - Technical MarketingI worked on the development of data processing and cleaning techniques to be used for evaluating datasets for quality and fitness for use. Some of these techniques are used for automating the standardization of datasets to be used in complex spatial analysis. I worked on a number of demonstrations and proof of concepts on a client-to-client basis in order to market many of the technical aspects of the ArcGIS Platform - particularly aspects related to advanced modelling procedures utilising the Python scripting environment. In this time I had the opportunity to present a seminar on Geocoding, another on Spatial Statistics and have helped develop course material on both of these topics.March 2015 - May 2015 - Professional ServicesI headed up the development, refinement & testing of the core Esri geocoding engine and developed an in-house automated address normalization and parsing algorithm. In this time I also defined cross-validation data processes to evaluate the quality of the outputs of a geocoding engine in a statistically robust way. Show less

• 

  Business science corporation

  Sept 2017 - Jun 2018

  Business consultant

  Development of a spatially constrained clustering algorithm implementation for commercial territory design that produces homogenous territories based on business infrastructure, demographics, customer behaviour and accessibility to assist in decision support and focussed strategic planning.Business development utilising targeted campaign design to drive additional revenue streams.The creation of various spatial tools to uncover hidden patterns in consumer behaviour & market share as well as the development of a recommendation engine for intelligent offer assignment.Development of many dynamic reporting mechanisms including custom components there in to inform decision making across verticals within the organisation. Show less

• 

  Harambee youth employment accelerator

  Jul 2018 - Dec 2019

  Data scientist

  The role entailed many aspects related to the data space within the business, including:- Evaluating platform architecture & pipelines for opportunities for cost reductions- Debugging issues with interdependent complex micro-service architecture- Overall efficiency & accuracy improvements across the platform data stack- Considerable data mining- Input on product design, platform architectural design & experimental design- Technical training- Technical support to members of the data team regarding best practices & optimizations they can leverage in their workProjects I worked on in this time:- Predictive modelling of behaviour scores- Deterministic engagement model to measure and improve network health- Similarity scoring matching engine for candidate recommendation configurable by business users- Model to test ethical assumptions across behavioral scores, identify biases in existing constructs & recommend viable corrections where possible- Clustering pipeline leveraging manifold techniques on exclusively categorical data to identify candidate archetypes- Distributed, scalable spatial-matching engine modelling the South African taxi network which provides accurate field-verified values for travel distance, time and cost for a commuter- Spatial sampling technique & ML-derived heuristics for an invite engine that reduced overall cost of invites & resulted in stable pitch rates for many months significantly improving the candidate experience- Address parsing & geocoding algorithm that achieves a higher spatial precision and accuracy than Google’s Geocoding API- Force-directed simulation of the Harambee employer network to model complex employ-employer relationships through the employment journeys of Harambee candidates- Visual testing suite to evaluate the impact changes to the matching engine have on the candidate pool across job families & regions- NN to translate address descriptions into coordinates for informal addresses within 2km accuracy Show less

• 

  Discovery limited

  Jan 2020 - Nov 2020

  Head of data science for vitality

  A few interesting projects I've personally worked on while here:- Bayesian LSTM-Attention based neural network architecture with temporal masking including a custom temporal SHAP implementation to help explain how the neural network was making its decisions by highlighting what it considers important to focus on at different points in the available timeline.- Development of a Monte Carlo simulation framework that models all possible points earning that could affect status with key behavioural dynamics built in based on past experience. - Development of a model used to understand how member physical activity changed as a result of lockdown- Design of a causal inference methodology for uplift modelling and to monitor campaign effectiveness across different channels - including causal discovery, the exploration of meta-learners and more traditional techniques like propensity score stratification- Thought leadership on how best to configure the scoring algorithm used in the COVID Alert SA Exposure Notification app- Development of a benefit registration-utilisation index used to monitor the health of each benefit for overall product design optimisation- Epidemiological modelling of Covid-19 spread in South Africa- Lead a team that produced a vulnerability score based on how likely someone is to have an adverse reaction to Covid-19 based on their medical claims history- Analytics to uncover suspicious patterns and the development of simple heuristics that can be used to indicate fraudulent activity at a member, broker and partner level- Development of automated reports that unpack rich cross-company detail for cohorts of members the marketing team is interested in targeting for different campaigns to help craft personalized messaging- Cross-company analytics looking at highly integrated members with different products across Discovery exploring links between what members do in the different product houses and how it affects engagement elsewhere Show less

• 

  Mercedes-benz south africa

  Dec 2020 - May 2022

  Solution architect (data & ai)

  Design & implementation of an AutoML framework for vision applications (classification, object detection & semantic segmentation) within a manufacturing context. End-users are factory floor workers with no knowledge of ML. A model is trained in the cloud and deployed to an IoT device in the factory.-Successfully delivered 10 different stations (7 object detection, 1 classification, 2 semantic segmentation) running live in production. -Extensive collaboration in a team context with my main focus being the ML aspects of the solution-Close collaboration with NVIDIA to enable an Non-Max Suppression module in TensorRT 8-Close collaboration with Microsoft in developing end-to-end tests to act as quality gates that evaluate all components of the solution running on an edge device on PRs to the edge device component repos-Horizontal scaling for training in Azure (Databricks & Data Factory) to train models independently in parallel-Centralized logging to Application Insights for the support team to query-Extensive experimentation and research to identify suitable model architectures and design choices that generalize and scale well across use casesTraining optimizations:-Transfer learning with fine tuning to train expert models with limited data in a short amount of time-Custom GradCAM++ implementation to work with a feature pyramid-Model quantization and conversion to TensorRT engine for fast and efficient inference on edge devices-MLflow logging across training runs-Automated learning rate range test to force fast convergence with a one-cycle policy-Auto-balancing of imbalanced datasets-Synchronous model evaluation during training for early stopping and real-time feedback of model performance during training-Custom boundary aware loss function for semantic segmentation to preserve geometries faithfully-Dynamic augmentation at training time to significantly reduce the annotation burden on the end-user-Custom ONNX graph re-writes Show less

• 

  Sand technologies

  Jun 2022 - Mar 2024

  - Grew and led a cross-functional team of data scientists, fostering a culture of innovation, collaboration, and continuous improvement, including mentorship of technical leads and juniors alike.- Crafted and managed the data science strategy, drove innovation, identified growth opportunities, and represented the Science Capability to stakeholders.- Established high-quality data science standards, reviewed methodologies, provided technical guidance, and ensured adherence to best practices.- Collaborated with clients and internal stakeholders to build and maintain data products, shaping solution architectures with AI-specific considerations.- Ensured the delivery of high-quality, scalable data-driven solutions, and oversaw the implementation of initiatives with strong adherence to industry standards.- Provided industry-leading insights, remained abreast of the latest methodologies, technologies, and trends in data science. Show less My main focus was establishing and formalising the R&D function across Sand Tech (formerly ExploreAI). Among other things, this entailed developing a framework for prioritising low-risk, high impact R&D which broadly covered two modes: leveraging hundreds of junior interns at a time from the ExploreAI digital skills academy and leveraging smaller, more specialised teams with senior data scientists and engineers. A key factor that makes this possible is the establishment of an opinionated process that helps distill big ideas into tangible, value-driving solutions that can be appropriately sized with a high chance of success.In addition to this system-level work, I also:- Provided deep technical guidance across a broad range of domains to R&D teams and Enterprise consulting teams throughout the data science lifecycle, personally contributing code to projects leveraging techniques like computer vision & multi-agent reinforcement learning.- Conceptualised, developed and maintained a remote work interpersonal tool that fosters a stronger sense of community across our globally distributed employee-base.- Conducted interviews for potential Explorers at various stages of their career journey. Show less

  • Head of Data Science

    May 2023 - Mar 2024

  • Technical Lead: Research & Development

    Jun 2022 - Mar 2024
• 

  Getsmarter

  Feb 2023 - Apr 2023

  Assessor - oxford artificial intelligence programme

  Responsible for grading assignments and providing individualised feedback for exec-level participants.

• 

  Ai sweden

  Apr 2024 - Nov 2024

  Ai engineer

  - Part of the Natural Language Understanding (NLU) Lab.- Developed next-generation multimodal models for Nordic languages.- Focused on creating tooling to efficiently curate and refine high-quality web-scale datasets on HPC clusters that integrate interleaved image, audio & text modalities for medium-to-low resource language contexts.- Designed, developed and orchestrated scalable end-to-end processes for dataset creation, training, and validation of native speech-to-speech models leveraging multi-node GPU parallelism including interactive web-based demonstrations of model capabilities. Show less

• 

  Imagimob

  Dec 2024 - now

  Senior ai engineer

  - Developed an explainable AI tool to assess the quality of classification models through detailed interactive analysis of embeddings as part of a steerable data cleaning feedback loop to improve the robustness of decision boundaries of downstream models.- Prototyped an auto labelling tool to create precise ground truth data for radar segmentation use cases at scale leveraging other aligned sensors.- Leading a project to develop accurate, configurable Edge AI for consumer radar systems.