Looker EngProd

Serverless Containers

• Optimized model loading for GPU-based AI workloads using OverlayFS (~200ms reduction),

eBPF (~15% latency reduction), and model snapshotting

• Analyze performance of competing internal Serverless products by running SQL queries and building Python notebooks to analyze overall latency distributions by customer

• Help use this knowledge to build a roadmap for migrating from one product to the next

• Target analysis data to identify specific performance improvements for the slower product

• Implement a performance improvement that utilizes the new eBPF Linux kernel plugin technology to detect when an app is listening on a TCP port

• Automate updates to Linux kernels used to power Cloud apps and integrate with experiment-based data analysis

Serverless Experiments

• Design a better API for running experiments (A/B testing) on Cloud backend features

• Improve performance of A/B testing code across all Google Cloud Serverless production code and reducing CPU usage by 33% (6% of all CPU -> 4% of CPU) by cleverly caching and reloading when needed

• Design/implement new experiment types that allow tracing data across multiple services and coordinated experiments across multiple backend services

• Migrate to new standard experimentation platform, give presentations to other teams and teach engineers across Serverless how to use the new platform

CI Infra (May 2020-Jan 2022)

• Designed and implemented a way to dynamically schedule jobs during execution of CI. Used by developers to significantly reduce AWS cost and reduce time to merge code.

• Created an Airflow metrics pipeline to collect data on all unit and integration tests at Airbnb. Used to measure impact of integration testing and reduce defects.

• Implemented system to identify and notify teams for each flaky unit and integration test via Slack and JIRA

• Designed and implemented a UI with Figma and React to search for unit tests and view past successes/failures

• Integrated with Bazel to allow disabling of flaky unit tests and consumption of test results from Bazel Build Event Protocol

Tax Platform (Jan. 2019 - Apr 2020)

• Worked on improving scalability and performance of a platform to calculate taxes

• Reduced latency by 50% on some critical paths and developed model to gracefully respond to extremely high load

• Designed and implemented a framework to schedule and execute cron integration tests

TA for Algorithms, Software Lab, and Software Design

• Graded assignments

• Wrote exam questions

• Led discussion sections and went over exams with students

• Tutored students 1:1/ Held office hours

Trust Tools

• Developed a backend for consuming and storing different types of messages written by agents working on fraud cases

• Integrated backend system with multiple teams