I architect distributed systems that hold up at 3 AM.
Two careers, one instinct: design for the load before you draw the details. I build event-driven architectures, real-time platforms, and the developer tooling that holds them together.
§02 — LOAD-BEARING NUMBERS
SOURCE: PRODUCTION TELEMETRY · 2024–202647-minute pipeline → under 8 minutes via per-app change detection + JIB.
EKS 1.32 + Karpenter v1, solo DevOps owner across a 7-person team.
EventHorizon, the pattern that crystallized after four companies of the same problem.
§03 — FIELD WORK
Selected Case Studies
Pipeline Platform: A Config-Driven Document Processing System
A multi-service platform that replaced a 534-processor monolith with a config-driven DAG, decoupled sub-processors over RabbitMQ, an operator SPA that owns no mutations, and a standalone JWT issuer designed for zero-downtime cutover.
Pipeline Orchestrator: One DAG to Replace 534 Processors
How a single Micronaut orchestrator replaced 534 concrete processor classes at a law-tech startup, and the subtle invariants (terminal state detection, optimistic-lock retry, transaction boundaries) that made it safe in production.
Pipeline Workers: Ten Generic Stages That Coexist With a Legacy Schema
How ten generic pipeline stages replaced court-specific logic in a document-processing pipeline, talked to the orchestrator over RabbitMQ, and shared a database with the legacy system without ever touching its schema.
Pipeline Dashboard: A Thin SPA and the API-Perimeter Lesson
Building an operator dashboard for a config-driven pipeline platform (read-only at launch, then a deliberate write surface for court config and patch-and-replay), plus the refresh-on-401 client middleware and the architectural lesson hiding in 'just add an nginx location block'.
Pipeline Auth: A Standalone JWT Issuer With a Zero-Downtime Cutover
How I designed a standalone JWT issuer for a pipeline platform, the dual-controller alias pattern that let pre-existing write surfaces keep working through the cutover, and the long tail of small infrastructure gotchas that cost more time than the code.
MCP Bridge: Shipping the Data Plane, Deferring the Framework
How I scaffolded an MCP server for a pipeline platform while the underlying micronaut-mcp library was still moving across three incompatible annotation styles and three JVM versions, and why the data plane shipped before the framework wiring did.
EventHorizon: The Ultimate Data Detection Engine
Designing a generic, pluggable anomaly detection platform using Apache Flink, Kafka, Micronaut, and ML/LLM integration. The culmination of building similar systems in Ruby, Java, Python, and TypeScript across multiple industries.
CI/CD Pipeline Optimization
How I restructured a monolith's build system and CI pipeline: replacing Dockerfile builds with JIB, adding per-app change detection, and cutting pipeline times by 83%.
EKS Cluster Migrations
Upgrading EKS clusters to v1.32 and migrating Karpenter to v1 across 100+ production services with zero downtime, all as a solo DevOps engineer racing to avoid extended support charges.
Performance Optimization
47-minute CI pipelines down to under 8. JIB + change detection + a structural untangle of 32 Gradle projects.
System Architecture
Real-time anomaly detection on Flink + Kafka. Same backbone, four industries.
Cloud Infrastructure
EKS clusters upgraded to 1.32 across 100+ services. Solo, deadline pressure, zero downtime.
Explorations
Agent orchestration, shared context layers, the next layer of developer tooling.