Supermarket self-service checkout Image Classifier

Deep learning neural network image classifier for the automated identification of food products. Included the development of a custom neural network for image classification (deployable to mobile devices), front end user interface and peripheral integration (camera, scale and printer).

neural-network-01-01.png

text_identification_neural_net.png

Remittance and INvoice Image recognition and machine learning

End-to-end integrated pipeline and machine learning solution for extracting key fields from scanned remittances and invoices, calculating line totals, matching them with bank statements and automating reconcilement and follow-up communications with customers.


Fraud detection and risk management system

Payment and claims fraud system using machine learning that reveals instances of fraud and recurring late payments in datasets containing millions of transactions and thousands of customers. This system is also coupled with a custom built automated notice and rectification system.

decision_tree.png

wood_xray.jpg

Wood identification and classification through image recognition and machine learning

The classic convolutional neural network approach used for image and object recognition repurposed to identify and classify different species of wood samples based on X-ray and microscopy images of their internal structure. Based on visual indicators, this method can identify the species of tree, age, originating location of a wood sample, and potentially the physical characteristics and quality of the wood.


Document extraction using machine learning

Image recognition and document extraction solution that extracts key fields from unstructured insurance data sources (PDFs, word documents and text files). The system uses computer vision techniques to process documents, machine learning to identify content types, and extracts the classified content into a structured format. The entire process is callable via a REST API.

 
multiple_file_types3.jpg

trading.jpeg

Automated online data extraction for market pricing and benchmarking

Developed a series of interactive web bots which extract data from online sources to provide insight into market pricing at policy level during acquisition and renewal. This included benchmarking pricing against competitors and discovers more about customer behaviour by investigating switching trends across the market.


social-media2.png

Predicting credit defaults with machine learning AND PUBLICLY AVAILABLE DATA

Support vector machine that predicted the likelihood of a company default. This model trains itself over time using information obtained during the application period and ongoing data collection collected through publicly available sources (news articles, ASIC and social media).


data_visualisation.jpg

Real time analytics and visualisation

Developed a series of real time analytics reports and interactive visualisations. This included connecting dozens of different data sources ranging from data warehouses to Excel spreadsheets. This included the development of several interactive dashboards available through the browser using a combination of Tableau and D3.


machine-reading.png

Machine learning to predict salesperson performance

Multivariate regression to predict the performance of individual salespeople based on their personal history, company performance as a group and naturally occurring workload cycles.


ruby_code.jpg

Application development and hosting

Developed a white labelled online loan application form and streamlined automation of the approval process. This included the enhancement of existing web applications, with the hosting and management of the existing AWS, Ruby on Rails and Drupal environment, backed by a PostgreSQL database.


machine-reading.png

Automated regression testing and system monitoring

A range of tools that identifies system faults and detects application downtime. This includes the automated testing of new deployments prior to release into production.


220px-Rayleigh-Taylor_instability.jpg

Finite Difference Fluid Flow Simulation

Upgrade of historical codebase from serial FORTRAN application to a massively parallel C++ program deployable on supercomputers. The resulting application was capable of conducting fluid flow simulations to upscale petrophysical properties from pore to core scales on 3D grids containing over 100 million cells.