Tony Vuong

At Amazon, I led architecture and productionization of the Amazon Bedrock Agents runtime, covering orchestration, model and tool execution, deployment, and reliability. I architected a parallel execution layer that reduced workflow latency by 60% in production.

I also designed a distributed agent execution framework with worker pools and SQS for concurrent, multi-tenant workloads. I added retries, circuit breakers, and automated canary deployments, reducing rollback recovery from 30 minutes to under five.

For Amazon Stores search ranking, I worked across retrieval, feature pipelines, evaluation, and model serving. I architected a multi-objective Learning-to-Rank framework and rebuilt offline evaluation, cutting experiment turnaround from four weeks to two.

At Facebook, I built and scaled machine learning infrastructure for Feed ranking. I migrated feature pipelines to near-real-time Kafka streaming, reducing feature freshness from 24 hours to under five minutes, and contributed to evaluation, model serving, and production reliability.