KANGZE JIA

Led CI team to cut E2E PR merge time from 50 to 20 minutes at p50 and from 120 to 67 minutes at p90, while daily merged PR volume grew from 1000 to 3000.

• Redesigned merge queue recovery with N-sect failed-test-only reruns. Built probability-guided speculation strategy to prioritize most useful alternative execution paths.
• Improved infrastructure reliability through zram, secondary disk mount, CI node dirty-shutdown elimination, tmpfs etc..
• Advanced test reliability by piloting LLM-assisted unstable test detection and fixing.

Led the AWS re:Invent Redshift Multi-AZ deployment project, delivering a highly available, mission-critical data warehouse for enterprise workloads.

• Enabled active-active operations across two Availability Zones for both reads and writes, doubling performance compared to Single-AZ deployments.
• Raised Redshift SLA to 99.99% with RTO < 1 minute and RPO = 0.
• Implemented failure simulation to validate resiliency and recovery.

Led the refactor of Redshift Capacity Management Service which does capacity procurement, host creation, pooling, and fulfillment.

• Implemented comprehensive throttling which lowered ICE risk for critical requests (e.g. Resume) without sacrificing utilization.
• Simplified architecture by removing pool hydration prediction logic, avoiding overlap with procurement predictions.
• Reduced fulfillment latency by maximizing hydration of available capacity into pool. P90: from 5s to 800ms, P99 from 200s to 4s.

• Led the design and implementation of a next-gen fraud evaluation platform using the E-GMRA framework (Event, Gather, Model, Rule, Action) to combat AWS Signup fraud-boosting capture rate by 16% (to 61.2%).
• Enabled 80%+ of new fraud evaluations with zero code changes, allowing rapid updates to counter evolving fraud patterns.
• Optimized variable gathering with Akka Streams, reducing dependency calls, eliminating redundancy.
• Designed for extensibility, supporting additional use cases such as EC2 Instance Launch events and Risk Score generation.

• Implemented data gatherers to enhance fraud detection via ML, Scaled manual investigation workflows by shifting long polling to an event driven architecture, Integrated Designed Party Screening into AWS Signup to ensure the legal compliance.