Working on the ML team

Landscape and travel photography + videography.

Working with the Computer Architecture and Programming Abstractions group, led by Adrian Sampson.

Formalized operational semantics and proved type soundness for Dahlia, a programming language for designing hardware accelerators with abstractions that guarantee predictable hardware realizability after type checking. Published in Proceedings of ACM SIGPLAN PLDI 2020.

Working with the ML Platform team on MLflow, a framework for managing the end-to-end machine learning lifecycle.

• Created an automatic logging integration framework with support for Keras and TensorFlow, providing out-of-the-box metric+parameter tracking for model training sessions with zero code change, greatly improving MLflow's ease of use.

• Made significant performance improvements to the Databricks MLflow backend by implementing SQLBatch logic for query processing in bulk.

• Led an initiative to improve interaction with the open-source community by revamping policies on issues and pull requests on GitHub, and organizing an issue bash event to close out long-standing bug reports.

At Dell, I created a distributed tracing system for EMC's Elastic Cloud Storage platform to provide rich information about individual requests as they propagate across services in the distributed storage system, down to the method level, for better debugging and bottleneck identification.

This involved designing an interface for trace propagation between threads and across remote procedure calls. Along with this, I wrote an extension to the javac compiler to automatically generate Java bytecode through AST manipulation at compile time.