Projects

The networking layer for IPv6 adoption across ~700,000 public RDS databases (18.5% of the fleet) in 30+ AWS regions. I engineered the end-to-end workflows on a split-horizon DNS architecture, so the same endpoint resolves correctly from inside and outside the VPC, over v4 and v6 alike.

Migration at this scale fails in the seams, so I built a validation suite covering nine critical failover and network conversion scenarios before any customer touched it. The launch landed cleanly: 200+ enterprise adoptions within the first 60 days.

On-call engineers were losing hours hunting through runbooks mid-incident. I architected a retrieval-augmented diagnostic tool on Amazon Bedrock that indexes 10,000+ internal technical documents in OpenSearch for semantic search. Ask it about a failure mode and the relevant runbook sections come back in seconds.

Result: incident investigation time down 74%. The fastest way to fix an outage is to stop searching for the document that explains it.

Lambda's networking logic was tangled into a larger service, slowing every cold start and every deploy. I led the decoupling: migrated it into a dedicated gRPC microservice with parallel initialization.

Result: $3M in annual infrastructure savings and cold-boot times down 180 seconds. The same work pushed DNS availability for Lambda non-VPC environments from 94% to 97%.

When an RDS Multi-AZ database fails over, the standby takes over fast, but customers can't reach it until DNS propagates. Every second of that propagation window is downtime a customer can see. I'm leading the networking work to shrink it.

Projected impact: DNS propagation time during failover down 70%, across millions of Multi-AZ customer databases. Less time waiting on DNS means less visible downtime on every failover, planned or not.