PerOmnia LLC · Chicago, Illinois
Contract backend, infrastructure and integration engineering — queues, webhooks, deploys, compliance, and the services that have to still be running at three in the morning. Available now.
Twilio end to end — A2P 10DLC brand and campaign registration, delivery failures, quiet-hours rules, STOP and HELP handling, consent logging. A small, painful specialty most teams discover the week their messages stop arriving.
A service that won't start, a deploy that broke, a unit crash-looping with an exit code nobody recognises. I find the cause rather than restarting it and hoping — and I check what is genuinely running in production before changing anything.
Canonical URLs, sitemaps, robots, Open Graph images and structured data wired into the App Router — generated from your content rather than pasted beside it and left to drift out of sync.
Container builds, image delivery, ingress, secrets and rollout mechanics for teams who need it to work reliably without standing up a platform group to babysit it.
Twilio · Node
Postgres · BullMQ
A form takes a lead and a text goes out — except the moment real money is involved, “send a text” becomes consent law, carrier registration and time zones.
RESULT Messages queue in ~30ms and either send inside the legal window or hold to 08:05 local. Retries are idempotent — a lead cannot be texted twice by a redelivered job.
Next.js 16
App Router · TypeScript
The site ranked for nothing and had no machine-readable description of what it sold. Rebuilt so search engines and AI assistants read the same facts a visitor does.
metadataBase with per-route canonical URLs, plus generated robots.ts and sitemap.ts.next/og, so social cards never go stale.SHIPPED 12 files, 618 insertions, one commit. VERIFIED tsc clean, production build passes, served HTML carries canonical, OG and Twitter tags plus exactly one ld+json graph whose tier prices match the page.
systemd · Celery
Debian
Scheduled jobs weren't running. The service reported itself failed and restarted
forever, exiting 226/NAMESPACE — an error raised before any
application code runs, so the logs were empty.
ReadWritePaths=/tmp against a private tmp namespace.StateDirectory=, with CPU and memory caps because the host is shared with a latency-sensitive service.RESULT Scheduler running clean, resource-capped on a shared host, and the fix committed to the copy production genuinely deploys from.
Kubernetes · containerd
Cloudflare Tunnel
A single-node cluster running eight independent services — Telegram bots, a Next.js API with a worker, scheduled content jobs, static sites and a Base L2 subscription layer.
docker build is invisible to this kubelet — a failure mode that silently keeps serving the previous version.RUNNING NOW 8 services, 8 namespaces, single node.
Something is broken and revenue is at risk. Hourly, start same day, diagnosis before changes.
Defined scope, quoted up front. SEO passes, integrations and pipelines usually land here.
I take the backend piece under your name. You keep the client relationship.
I work in your repository, on your branch, with commits you can read. If a job turns out smaller than quoted, I say so and bill the smaller number.
Send me the problem, not a job description.
If it isn't something I should take, I'll tell you that and point you somewhere better.
Code samples on request — most of what I ship lives in private client repositories.