Jesse Moore

The goal for this project from the outset was to predict the outcome of hockey games and attempt to find systematic errors in betting odds by traditional providers such as Bet365 and Pinnacle.

The system began to approach parity in capability with the best provider in January 2019, and surpass it in February.

Put into production in March 2019 and successfully traded throughout the following 12 months, the system has returned over 900% to investors.

Sigmai’s proprietary NLP technology enabled our customers to be the first to act on breaking news and allowed us to build an unbelievable archive of corporate history.

Newsful was a demonstration of that technology, and was shortlisted for a CODiE Award in the Best Business Intelligence Reporting & Analytics category.

Lead author of a proprietary end-to-end deep-learning-based NLP system for classifying text.

Achieved state-of-the-art results for entity-based classification (classifying text as it relates to a specific entity in the text).

Leveraging the growing ability to develop world-class natural language processing (NLP) algorithms using transfer learning in a fraction of time, I wanted to create the simplest classifier I could using the fewest lines of code possible. The goal of this classifier was to have it perform without any customization required by the user aside from testing various hidden layer sizes.

I explain "Using BERT To Accelerate Natural Language Processing". Data collection is burdensome, time-consuming, expensive, and the number one limiting factor for successful NLP projects. Preparing data, building resilient pipelines, making choices amongst hundreds of potential preparation options, and getting "model ready" can easily take months of effort even with talented machine learning engineers. Finally, training and optimizing deep learning models require a combination of intuitive understanding, technical expertise, and an ability to stick with a problem. Google's BERT (Bidirectional encoder representations from transformers) form transfer learning that requires far less data and training time. Building world-class models become the possibility of any data scientist, rather than solely the domain of large companies with large budgets.