Guilherme Morais

A prediction model to classify hypertensive patients according to their risk of developing severe complications, such as stroke and heart failure. The model was trained on data collected by community health workers in Brazil during routine household visits. The goal was to leverage the current structure of primary care and be readily available for scalability in the country. It was endorsed by local physicians and set to be validated with a randomized controlled trial.

Real-time optimization of pricing strategies based on consolidated market data and observed costs. The algorithm was embedded into a dashboard served on the cloud and presented features for simulating key business metrics within different scenarios to enable stakeholders' decisions.

A Python-based pipeline communicating with cloud services (Amazon S3, Amazon Athena, and AWS Glue) to consolidate incoming external data into the company's data lake. The pipeline could be triggered by identifying new data available and then run all processes autonomously. For each step performed, relevant logs would also be stored for tracking and debugging.

A Python package with refactored code for data science in healthcare. It provided modules to facilitate the processing, visualization, and modeling of data structures commonly used within the healthcare industry. It enabled a team of several data scientists to quickly and reliably run exploratory data analysis, model training, validation, and visualization within hours instead of weeks.

A Kubernetes-based platform implemented on AWS with a common endpoint for the data science team. Users could allocate the necessary resources on a base instance and instantiate on-demand clusters for parallel computing and GPU computing. The instances were created upon containers of a common Docker image to guarantee reproducibility of the environment by team members.