Senior Software Engineer · Java · AWS · reliable systems at scale · tools for AI coding agents
Worked with: John Deere · AWS · SpaceX Starlink (Deere partnership) · Mobifone · open source at Apple, Stanford and CMU
John Deere · 2026 · Team program; I built the equipment platform part.
Shipped: three services running in a second AWS region.
- Aurora Global Database with AWS JDBC Wrapper failover, switchover scripts and AWS FIS fault tests
- Cross-region SQS listener takeover in a shared library, used by two services
- Valkey global datastore, Route 53 health checks, blue/green deploys with rollback gates
Links: Public failover lab · Live architecture map · Floci #6213: Aurora via AWS JDBC Wrapper
John Deere · 2025–2026 · Java 21, Hibernate, Valkey, Datadog.
Shipped (Jul to Oct 2026): API p95 latency about 43% lower while traffic grew about 34%. CPU per request about 50% lower. Message service memory use 34% → 15%, with 23% fewer tasks. Gains coincided with the changes below (monthly data).
- Statement-count test harness; removed redundant
DISTINCTand Hibernate hydration - Authorization cache on Valkey, per-type
JAXBContextcache (about 15% less CPU at fix time), JVM heap sized from task memory - All services moved to JDK 21, JUnit 5 and AWS SDK v2
- 2024: Deere x SpaceX Starlink integration, the satellite registration flow for connected machines
Public context (employer code is private): Deere calls Operations Center "the digital twin of a farm" and aims for 1.5M connected machines and 500M engaged acres by 2026 · Deere and SpaceX Starlink partnership (2024)
Architecture analysis library and MCP server for AI coding agents.
Shipped: 23 merged pull requests and a GitHub Action.
- Kotlin parser, polyglot parsing, async analysis pipeline
- Architecture drift reports in CI
Links: Repository · My merged PRs · arcade-agent-action
NHTSA VIN decoder at the edge, on Cloudflare Workers.
Shipped: live API, drop-in compatible with NHTSA vPIC, listed in public-apis.
- Decodes VINs offline; every data update is checked against the live API and NHTSA's own SQL
- OpenAPI 3.1 docs, batch page,
llms.txtfor agents
Links: Live site · API docs · Repository · public-apis #8432
| Project | Pull request | What changed |
|---|---|---|
| Apple ServiceTalk | #3677 | Keep demand across overlapping subscription switches |
| Stanford CoreNLP | #1530 | Remove an O(N²) copy in CoreLabel and ArrayCoreMap |
| CMU Fray | #446 | Report replay schedule divergence |
| Embabel Agent | #1832, #1812 | MCP end-to-end tests, RAG ordering fix |
| Floci | #1951, #6213 | Terraform RDS, RabbitMQ and Aurora JDBC compatibility |
| Underthesea | #1053 | Lazy-load the translation backend |
| Years | Where | Role |
|---|---|---|
| 2021 – now | John Deere | Senior Software Engineer (via Volt, 2021–2022) |
| 2011 – 2017 | FTL JSC | Deputy CEO and tech lead · high-transaction services for Mobifone |
| 2008 – 2020 | FTL, eXo Platform, Tuan123, Xvolve | Full-stack and tech lead |
Stack: Java 21 · Spring Boot · Hibernate · AWS (Aurora, SQS, SNS, ECS, Route 53) · Terraform · Datadog · OpenTelemetry · React · TypeScript · MCP
Education: M.S. Computer Science, Maharishi University of Management (2022) · B.S., Hanoi University of Technology (2010)