Software Engineer (Application Engineer III)
Benchmark Construction Service
TypeScript · Lambda · SNS/SQS · DynamoDB · AppSync
- Re-architected the serverless service that calculates the index benchmarks behind portfolio trading decisions, reducing production incidents from daily to fewer than 3 per year at a run cost under $1K/month. I now own its data model and design direction.
- Engineered idempotent, order-independent ingestion of 100M+ price, index, FX, and reference records/day with transactional batch writes, so consumers receive only complete datasets.
- Reduced MTTR from 5+ hours to under 1 hour with replay/redrive, dead-letter queue alerting, and Grafana, X-Ray, and OpenTelemetry observability.
- Delivered the rebuild in 4 months on a 3-developer team, with all infrastructure as code in CloudFormation, CI/CD through Bamboo and GitHub Actions, and no breaking changes to the GraphQL API.
Valuation Platform
Go, Java, Kotlin · Kinesis · AWS Managed Flink · Aurora PostgreSQL
- Co-designed, with the senior tech lead, a Kappa-architecture streaming platform for real-time multi-asset valuation and liquidity, now partially in production.
- Modeled the platform's star schema, now in production across 4 domains: benchmark construction, look-through exposure, country-of-risk classification, and portfolio liquidity.
- Redesigned the benchmark data model so a price update recalculates only the affected constituent instead of the whole benchmark (O(n) to O(1)), enabling real-time valuation.
- Built provider data ingestion into a multi-region change data capture (CDC) pipeline with up to one week of redrive for incident recovery.
- Engineered blue/green deployment at the data layer, enabling full-stack staging.
- Led technical direction for 6 developers building an AWS Managed Flink stream-processing prototype: defined work, mentored, and reviewed code.
Equity Data Lakehouse
Go · Kinesis · Firehose · EventBridge Pipes · AWS Glue · Apache Iceberg · S3
- Designed the lakehouse's multi-region architecture to survive a us-east-1 outage, choosing active/passive S3 cross-region replication and cutting multi-region run cost by more than half versus active/active.
- Built event-driven ETL in Go feeding an Apache Iceberg lakehouse that processes hundreds of GB/day for 100+ analytics, quant, and ML consumers, integrated with the internal data-quality governance API.