Case Study: Thomson Reuters accelerates NLP research and development with Amazon Web Services

Streamline and Standardize the Complete ML Lifecycle Using Amazon SageMaker with Thomson Reuters

Thomson Reuters, a global provider of legal, tax, and news intelligence, needed a faster and more secure way to research and build natural language processing and question-answering models. Its research team was iterating on BERT-based machine reading comprehension projects using large proprietary datasets, but on-premises infrastructure made experimentation slow, costly, and difficult to scale.

Amazon Web Services, using Amazon SageMaker, helped Thomson Reuters train, fine-tune, and deploy models with pay-as-you-go GPU resources and managed Spot Instances. The result was major efficiency gains: training costs were reduced by 40–50% on average, fine-tuning time dropped from many hours to under 1 hour on P3 instances, and pretraining was cut from an estimated several weeks to just a few days, enabling faster, more accurate search capabilities.

Thomson Reuters

Maria Apazoglou

Vice President of Ai/ml and Business Intelligence Platforms