Machine Learning Scientist
Current• Led the end-to-end development of a family of Transformer-based protein sequence VAE models, trained on the OAS dataset. This included hyperparameter tuning and optimization of training code to eliminate I/O bottlenecks, demonstrating my ability to select suitable architectures and improve efficiency in machine learning workflows.• Introduced modular management of training parameters using Hydra, enhancing the flexibility, traceability, and scalability of the model training process. Deployed an MLFlow Tracking Server and wrote custom callbacks to allow seamless run tracking within the training pipeline.• Engineered a fault-tolerant in silico antibody optimization pipeline that distributed parallelized jobs corresponding to 28 distinct experimental groups across a cluster of 48 EC2 instances. This was achieved using a combination of PyTorch, Ray, MLFlow, and Spark, highlighting my proficiency in distributed computing, cloud infrastructure, and machine learning at scale.• Spearheaded the design and implementation of modern security best-practices for a multi-account AWS Organization. This included the introduction of SAML-based Role-Based Access Control (RBAC) and standardized enforcement of Multi-Factor Authentication (MFA), demonstrating my commitment to cloud security and best practices.