Software & production systems engineer
Twenty years of designing backends, integrations, and automation that keep working after launch — when the inputs get messy, the load is real, and things start to fail. Not demos. Systems people can actually run.
I reach for AI where it genuinely improves the outcome, and lean on solid engineering for everything else: architecture, state, integration, and failure handling.
A typical shape of my work — integrations, queues, data, and review paths, with AI as one component rather than the whole system.
About
I'm a software engineer with more than two decades of experience designing and building systems across different industries. These days I focus on production systems and automation architecture — the work that makes software dependable inside real business environments.
I design for production from the start: state, failure handling, integration, and observability, with AI added only where it earns its place. I care less about a prototype that impresses a room and more about a system that can recover, explain itself, and keep running when the inputs stop being convenient.
I'm always interested in hard systems problems and good engineering conversations. If that sounds like your world, say hello.
Work
A product I'm building, and a few examples of the production systems I've worked on.
Design production systems visually, from first sketch to shared architecture.
A side project: map services, data flows, queues, providers, and operational boundaries on a realtime canvas built for builders and architects, then bring teams in when the design needs review.
Stabilized a multi-system automation that broke under real conditions — duplicate events, API timeouts, and manual reruns — by adding durable state, idempotent operations, and operator-visible exception handling so work could be stopped, resumed, or reviewed.
Made an AI-assisted document pipeline safe for production: separated extraction, deterministic rules, confidence scoring, human review, and audited write-back so output was checked before it touched downstream systems.
Untangled a tightly coupled set of services and queues with clear service boundaries, message contracts, and correlation IDs, so production incidents became traceable and recoverable instead of slow to isolate.
Made a scheduled and event-driven data pipeline dependable under real load with validation, error isolation, replay controls, and a recovery procedure operators could actually follow.