1. Made improvements to the tracking of mobile experiments and feature flags (parameter tracking) in order to scale for years of projected growth in mobile experiments and feature flags. Implemented a new cleanup policy, new oncall alerts, new partitioning strategy, and developed checksum/hashing-based deduplication of app version experiment and feature flag sets. Reduced memory usage of parameter tracking data in high QPS service by 40%.

2. The data stored in the Parameter Tracking system garnered interest from other projects. Mentored and advised engineers building tools for monitoring feature rollout, for server-driven user interface (SDUI), and for tracking mobile code changes. Helped implement some of the backend components of these tools.

3. Redesigned python package used at Uber in Jupyter notebooks to analyze experiments. Surveyed data scientists across the company to determine package needs and use cases. Refactor and validation jobs made the package reliably more compatible with versions of python > 3.7 and newer versions of numpy, pandas, scikit-learn and internal python packages. In the process, I coordinated with python teams to create python best practices at Uber.

4. Made improvements and reduced the round-trip time of API used for evaluating all mobile client feature flags and experiments. Collected data from multiple sources on round-trip API performance. Profiled the micro service handling request and found several bottlenecks. Resolved most bottlenecks and improved service latency by 20% and reduced payload size by 40%.

5. Rewrote several APIs in Uber's dynamic configuration service as part of migration from python to GO(golang).

6. Conducted a thorough investigation of an occasional data imbalance in A/B experiment results. Identified “clockdrift” as a possible root cause. Collaborated with security and infrastructure teams to resolve a company-wide issue with NTP clock synchronization, resulting in a clockdrift reduction of more than 90%.

1. Designed and implemented the workflow for tracking all features and experiments in every app version. Built several APIs and a pipeline from scratch that processed feature and experiment declarations in mobile code and updated the backend with backend representations of the features declared in code and storage of what features are in every app version’s client code. This allowed the experimentation platform to only evaluate the feature flags and experiments used in the app version’s client code, rather than all active experiments and feature flags (the prior behavior). This reduced the average number of experiments and feature flags evaluated at app launch by 80%, significantly improving service latency and round trip time. APIs and pipeline were highly fault-tolerant and idempotent to mitigate many points of failure in the CI/CD pipeline that the workflow consumed code changes from.

2. Developed a discovery feature that enabled Data Scientists to identify suitable cities and markets for testing specific hypotheses. Collaborated with Data Scientists and Product.

3. Implemented core components of a migration from Cassandra to MySQL. Designed and implemented schemas, and core components of the dual-write strategy during migration.

4. Refactored transformations in a Spark job written in scala that updated a hive table of experiment exposures. Updated job to use functional programming for transformations.

4. Refactored experiment analysis to allow reporting of results in the event of partial failure during experiment analysis.

5. Played a key role in the development and implementation of Thrift and GRPC based microservices as part of the redesign of A/B experimentation at Uber.

6. Developed APIs for searching for experiments in elastic search and updating experiments.

7. Migrated microservice from python2 to python3. In the process, I identified and resolved issues in the code and deployment process of the service.