Worked under Spanner EngProd team to adopt Jepsen, an open-sourced testing framework, onto Cloud Spanner by generating random workloads against a Spanner instance, recording the results and verifying the history of Spanner interactions
•
Designed a modularized structure that allows extensibility for new consistency benchmarks and customized workloads by decoupling generator types from workload types and history recording from history verification
•
Implemented bank and multi-register benchmarks based on the framework that can expose external consistency violations
•
Utilized the interpolation between Clojure and Java to adopt an open source history verifier, Knossos, to the framework
•
Implemented an optimized version of Wing-Gong Linearization Algorithm using Java to achieve better maintainability
•
Developed an interface that packages the framework into a Docker image and deploys it to scalable Kubernetes engines
Implemented a video database that improves querying time of the user given model by 7 times with 93% accuracy by constructing and choosing from 1.3 million cascades composing of 360 smaller classifiers
•
Shifted metric from accuracy to precision to eliminate 40% of false positives produced by original method
•
Processed 4GB of highly noisy csv files to train and fine-tune multiple LSTM classifiers that achieved 75% near accuracy and average absolute error of 6 dollars per 225,000 product
Developed platform responsive frontend pages and components using libraries including emotion.sh
•
Adapted to React Hooks API to restructure components declared as classes to functions with useState / useEffect to eliminate unnecessary code structure and achieve better code readability and reusability
•
Utilized Redux to keep track of complex state changes in components such as a search page
