Principal Machine Learning Engineer
Served as architect for and built out Montai's ML platform from scratch. Hired and managed a small team dedicated to this effort.The ML platform facilitated model experimentation that was reproducible, scalable and usable for a team of ML and comp-bio researchers. These models were used to nominate small molecules for in-vivo and in-vitro experiments in Montai's drug discovery program. They helped us find numerous hits that were later confirmed in lab.The platform included:- Support for data transformation, model training and evaluation, and analysis and report generation in a variety of languages and frameworks, including Python, pytorch, tensorflow/keras, and R.- Tracking of experiment metadata in a database (supporting multiple programming languages and both local and remote execution)- Environment management via Conda, Pip and Docker- Compute scalability via AWS Batch and a python workflow automation framework (Redun)- Automated image building and testing via a CI-CD pipeline- A software library shared within the team for common patterns and operations including testing, data transformations, S3 data manipulation and more.I cultivated a highly collaborative culture within the ML team, with engineers and scientists owning and reviewing code together. We rejected the notion of scientists "throwing a prototype over a wall" for engineers to deploy, instead working together to identify needs and develop rigorous solutions. As a result, our experimental throughput was very high and we achieved rapid refinement of model architectures.
