João Costa

• Built and productionized a Deep Neural Network architecture to extract multi-emotion intensity from Amazon's hourly fulfillment labor force's feedback, matching state-of-the-art performance. • Deployed a configurable SageMaker Pipeline to extract stratified random samples from over 400k records in several Redshift tables; and form Best-Worst Scaling examples. • Labelled the previous examples in Amazon Ground Truth with a custom UI to validate the emotion intensity model on the target data.

Under the project MATISSE - A machine learning-based forecasting system for shellfish safety, with the financial aid provided by Fundação para a Ciência e Tecnologia (FCT). The project is comprised of a multidisciplinary team from five institutions: DGRM, IPMA, IT, INESC-ID, and MARETEC. The work plan involved the implementation and training deep learning models for feature extraction (representation learning) of remotely-sensed satellite images through the Copernicus program.

• Graded a score of 18 out of 20. • Trained and analysed several Machine Learning models to predict the length of stay of inpatients in a hospital, achieving a 29.26% performance improvement over the baseline. • Created a python-based analytics dashboard for hospital data, connected to a SQL Server database.