From a ticket to something that runs in production — the steps I follow, what I hold to, and the projects where you can check.
The same sequence whether it is a defense subsystem, a client site or a weekend tool. Each step leaves something behind.
Read the ticket or the spec until I can write its acceptance criteria. In the defense work that was a formal SRS; elsewhere a short design note.
Fix the seam first — the API contract, the interface, the message shape. What sits behind it can change; the consumer shouldn't have to.
Build one feature end to end — endpoint, handler, data, UI — in its own folder, not one layer at a time.
Tests at the level that matters: unit for logic, integration against a real database, a benchmark when speed is the question.
A reviewed pull request, CI, and a deploy that checks health and rolls itself back.
Monitoring, backups proven by restore, and a written note on what changed and why.
Example API is one OpenAPI contract with ten implementations behind it, and a black-box conformance suite each of them has to pass before it is measured. Orbital Ops draws the same line between the propagation worker and the globe.
webplatform is ~30 modules, each owning its schema and talking to the others only through contracts and a transactional outbox. Order API is the same take-home built twice — layered and vertical-slice — so the trade-off is shown rather than argued.
No abstraction before its second real use. For my homelab I turned down a query-cache, codegen and form-library stack as weight a solo system doesn't need; this site runs on three runtime dependencies.
Benchmarks with pinned versions, warm-up windows and tail latency instead of averages. Where a number on this site can be derived from data, it is — the project count, the languages, the shipped-lately line on Now.
I run my own homelab: 19 services, health-gated deploys with automatic rollback, a staging tier, nightly backups that have been restored to prove they work. Being on call for my own code changes how I write it in the first place.
Two years in a closed military environment: system analysis, SRS and SDD documents, review before merge, technical presentations in English to international partners. I hold NATO Secret and Czech Secret clearances.
Flat surfaces, hairlines, one accent used as ink. Decoration has to earn its place — on this site the only thing allowed to glow is the tree.
I use AI agents every day and built claude-kit to make them follow a repository's rules: routing guards, risk gates, review passes. The commits stay mine; AI help is disclosed once per project.
The graph should name whoever's accountable. AI help gets one honest line per project, not smeared across the history.
If this is how you want work done on your team — or on your project — write to me.