KEJO SHIEH

AI-Powered Data Lineage Graph Tool to Trace 1,200+ Datalake Tables

• Designed and implemented an internal tool using the Model Context Protocol to auto-generate a lineage graph across 1,200+ AWS data lake tables.
• Solved a critical on-call challenge involving deeply nested dependencies; some tables required 40+ joins to derive, making incident tracing extremely difficult.
• Enabled fast impact analysis for the data team during outages, significantly reducing time to resolution, and preventing cascading failures.
• The tool became a core asset in the team's on-call toolkit and is now routinely used in production incident triage.

Reduced Functional Cookie Size by 67% on aws.amazon.com

• Led an urgent initiative to shrink two large functional cookies to 33% of their original size across a global multi-DNS architecture, handling 10M+ daily requests.
• Designed and evaluated 12 architectural solutions in under 3 days; drove executive alignment, and led implementation to completion in one month while strengthening monitoring and alarms to ensure safety and observability.

Advertising Cookie System for aws.amazon.com

• Architected and implemented a scalable advertising cookie infrastructure, processing 5M+ daily requests, launched with zero downtime.
• Applied the Factory Pattern to unify logic with functional cookies, achieving 60% code reuse across cookie generation and validation layers.
• Personally implemented major components, and delegated minor tasks to junior engineers with code reviews and guidance. Integrated pressure testing, canary rollout, and real-time monitoring for a smooth production launch.

Family-Friendly Ad Filter System for Prime Video

• Sole designer and primary implementer of an ad filter that excludes sensitive ads (e.g., medical, adult content) from family-friendly shows, flagged by the Prime Video Catalog.
• Optimized for less than 20 ms on top of the original 800 ms latency increase across 400 million daily ad requests; integrated with minimal disruption to existing services.
• Delegated minor tasks to SDE1, and ensured consistency through reviews. Gradual rollout with zero production regressions.

Launched Ad-Supported Prime Video Across North America and Europe

• Led the launch of Prime Video Ads Publisher Rules across NA and EU markets, collaborating with regional and product teams. Managed 10+ ad pacing and capping strategies to balance user experience with monetization.
• Conducted 20+ A/B experiments to validate algorithmic trade-offs, including user churn, engagement, and revenue.
• Identified & implemented algorithm optimization that boosted 2% of the whole Prime Video revenue, which is 20M USD annually.

• Maintained and improved server-side APIs for Nordstrom's online shopping platform serving over 1M annual users.
• Worked with AWS services including CloudWatch, CloudFormation, DynamoDB, EC2, and S3.
• Wrote unit and integration tests to ensure API robustness; introduced automated Postman scripts for QA.