Scaled up capacity and throughput of customizable data pipelines at a Series A startup by distributing computational and memory load over machine clusters.

• Developed implementations of modular platform components that operate on distributed datasets using Ray (by Anyscale).

• Increased (by over 100x) the size of datasets that can be feasibly supported on the platform from <10GB to >1TB.

Developed a Python API that facilitates the customization of computational chemistry pipelines running Amber (for molecular dynamics simulations) and TeraChem (for quantum mechanics computations) using modular components.

• Parallelized workloads by formulating a series of computations as nodes in a dependency graph.

• Used Argo Workflows on Kubernetes to orchestrate jobs in Docker containers.

• Yielded over 3x speedup on some workflows.

• Made computational chemistry tasks more robust to node errors or AWS spot instance termination.

Advanced molecule generation efforts.

• Designed a new interface and implemented 2 new molecule generation methods for manipulating and composing different combinations of substructures (fragments) when generating molecules.

• Integrated Unigen into a data pipeline that would automatically filter and use ML models to predict certain desired properties (e.g. ADMET).

Lead developer for Unigen, a platform for generating/enumerating molecules.

• Devised a unified, common interface for disparate generation methods. This facilitated the parallelization of molecule generation tasks across multiple tools and their integration into data pipelines.

• Implemented a suite of enumeration tools based on molecule generation algorithms from academic papers, open-source packages, etc.

• Created a Django web application allowing chemists to configure a selection of tools to run on a list of molecules.

Internship cancelled due to COVID-19.