I build tools at the boundary between Java bytecode, Spring Boot packaging, Linux memory behavior, and reproducible performance evidence.
My current project is JMOA, an evidence-gated, build-time JVM footprint optimization system. It does more than rewrite bytecode: it verifies the deployed artifact, proves runtime origins, measures process and cgroup memory, attributes regressions, and rejects results that do not survive frozen controls.
The exact accepted JMOA R41F fat-JAR deployment versus the documented strict no-JMOA B0 exploded-Boot deployment:
| Metric | Result |
|---|---|
| Process PSS | -15,241.5 KiB (-14.88 MiB) |
| Target-cgroup RAM | -17,033,216 B (-16.24 MiB) |
| Held-out consistency | 12/12 favorable blocks |
| Exact paired sign test | p = 0.00048828125 |
| Lifecycle CPU tradeoff | +14.71% median |
The campaign completed 81/81 sessions with no reused predecessor observations. The claim is packaging-inclusive and service-specific; the four-arm factorial showed packaging was the dominant measured contributor.
- JMOA 2.1 release
- Technical paper
- Source, architecture, and exact evidence
- Engineering case-study portfolio
- JVM bytecode and ASM transformation
- Maven plugins and build-time tooling
- Spring Boot fat-JAR/exploded deployment materialization
- Java 17+ runtime diagnostics: NMT, heap, metaspace, JIT, GC, and class loading
- Linux
smaps, PSS, cgroup v2, page-fault, swap, and reclaim analysis - controlled performance experiments, exact tests, and bootstrap inference
- failure-preserving automation and evidence-led release engineering
I am interested in JVM, Java platform, performance, infrastructure, and developer-tooling roles where deep debugging and careful experimental evidence matter.