Work I have built

Jimmy Hough Jr. · Swift, TypeScript, embedded, and AI-assisted development

Resume · Services & rates · [email protected] · github.com/jhoughjr

Every system on this page runs today, and I operate it. I designed each one and directed teams of Claude Code agents that wrote most of the code, under my specifications and my review. The numbers come from the repositories: lines of tracked code, test cases and commits.

Between July and September 2026 that came to about 119,000 lines of code and more than 3,600 tests, from one engineer. Each study ends with the package on the services page that buys the same kind of work.

A home app platform with the reliability of a cloud one

roost · Python and Bash · Dokku, nginx, Cloudflare Tunnel, restic · 220 tests · 238 commits

The problem. Small apps cost money every month on a cloud host, and each one needs HTTPS, deploys, backups and monitoring.

What I built. A platform on one Orange Pi 5 Plus at home, behind carrier-grade NAT, with no public IP and no port forwarding. A git push deploys an app to its own HTTPS address. One command takes a new app from nothing to a live URL. One dispatcher runs 39 tools: nightly encrypted backups with a restore runbook, disk reclaim, LAN certificates, and a reconcile job that repairs the box after a reset. Hosts read their secrets from a sealed store at boot, so no secret lives in a repository.

The result. Nineteen of my own apps run on it today, among them pulse, docs, watts and this resume.

The same work for you: Your own deploy platform, off the cloud bill, from $900.

Single sign-on and a file store for every app on a domain

vault · Swift, Hummingbird 2, swift-crypto · 194 tests · 86 commits

The problem. Each small app needs sign-in, permissions and a place to keep files, and writing them again for each app is slow and unsafe.

What I built. One sign-on service for the whole domain, with Apple, Google and GitHub accounts. A single signed session covers every app. Per-app authorization has five rule kinds, from “any signed-in person” to “a member of this GitHub organization”, plus named roles. nginx asks vault before it passes a request, so an app gets a gate with no code of its own. The same service speaks the Amazon S3 protocol, with SigV4 signatures, multipart upload and presigned links, so standard S3 tools and SDKs work against it. I first wrote it in Node, then ported it to Swift as a drop-in replacement.

The result. Live at vault.jimmyhoughjr.net. Apps declare their gate in one line of configuration.

The same work for you: the Premium tier of Your own deploy platform, $5,000, or A server-side Swift API, from $1,500.

A control plane for teams of AI agents

rookery · Swift, Hummingbird 2, PostgreSQL · 1,143 tests · 442 commits in three weeks

The problem. One AI coding agent is useful. Several working in parallel need roles, limits, a record of what each did, and a person who can watch and stop them.

What I built. A server that opens agent “seats” by role, streams every tool call each agent makes to a live board, and keeps registries of roles and shared skills. It has 94 HTTP routes, grouped by who may call them, and a test fails when a route is missing from the published OpenAPI contract. Agents reach it through an MCP server. The runtime boundary is one protocol, so the way an agent is started can change without touching the rest.

The result. Twelve agents measured clean in parallel on one subscription. I use it to run the agents that build the other systems on this page.

The same work for you: Claude Code agent workflows for your team, from $650.

One declaration for every app, database and scheduled job

hatchery · Swift, swift-argument-parser, swift-nio, OpenTofu · 824 tests · 164 commits

The problem. A platform drifts. Someone changes a port or a setting by hand, and nobody notices until a restart undoes it or an outage shows it.

What I built. A command-line tool and a local dashboard that declare every app, database and scheduled job, deploy them, and report each place where a machine no longer matches its declaration. Its audit has typed findings: a secret outside the secrets file, an undeclared database, an orphan role, a credential in a service file, a secret with no rotation, a port map that moved. It deploys to Dokku, DigitalOcean App Platform, AWS App Runner or Google Cloud Run. Cloning a stack decides key by key whether to copy, rewrite, mint new or refuse a value, and it never copies the source’s database. A 30-second watch keeps a history and alerts only when something gets worse.

The result. It declares the whole estate. Each change there is measured before and after against the declaration.

The same work for you: CI that tells the truth, from $500, or the Standard tier of Your own deploy platform, $2,500.

Status boards and fleet telemetry

statusgen and pulse · Python, JavaScript, nginx · 548 tests · 233 commits

The problem. A team wants to see build health and machine load at a glance, without a monitoring product to run and pay for.

What I built. statusgen turns a JSON file into a styled status board, with one shared renderer, a validated schema, and collectors that feed it live CI events. It deploys with one git push and has no build step. pulse shows live resource figures for every container and host, read from the Docker API through a read-only socket and from the kernel, with no npm dependencies. Macs push their own reports to it.

The result. Live at pulse.jimmyhoughjr.net.

The same work for you: the Standard tier of CI that tells the truth, $1,200.

A gig finder, from idea to live in eight days

gigs · Swift, Hummingbird 2 · 49 tests · 20 commits

The problem. Contract work is spread over many job boards, and most listings are full-time roles that do not fit.

What I built. A board that reads seven public feeds, only the ones that allow automated reads, and scores each listing against my skills. The scorer shows its reasons: skill matches, contract and remote bonuses, pay thresholds, and penalties. Each morning it pushes the strongest new contract gigs to my phone through Home Assistant. It drafts a pitch for a gig with the Claude API, on request.

The result. It runs behind the single sign-on above, and the history of each morning’s push survives restarts.

The same work for you: A server-side Swift API, from $1,500.

Apps for Apple platforms

videmus · Swift, SwiftUI, ARKit, Vision, Metal

An iPhone, iPad and Mac app that measures how far you sit from the screen and draws letters at their true angular size. Distance comes from one protocol with three tiers: TrueDepth face tracking, pupil landmarks calibrated against a standard ID card, or a distance you enter. Each reading carries its uncertainty. Through the camera it simulates blur, astigmatism, colour vision deficiency, cataract, macular degeneration and glaucoma, and on LiDAR devices each pixel blurs by its own distance.

lighthouse · Swift · 667 tests

An offline-first Bible study app. A build pipeline compiles five public-domain editions, about 163,000 verses, into SQLite.

The same work for you: A Swift, SwiftUI or Electron bug, fixed, from $300, or a quote for an app.

Production point of sale, for a client

Austin MacWorks · Swift (Vapor), Electron, React, TypeScript · December 2024 to now

I wrote about 650 of the 930 commits of Phoenix, the first point-of-sale client, and led the move to a desktop client in Electron and React. That client talks to the server through a typed SDK that is generated from the server’s OpenAPI contract. I built swift-pdf-builder, a pure-Swift PDF library for invoices and receipts, with 160 tests and green Linux CI on its first day. I added the backend’s CI: multi-architecture release builds, test reports and coverage, caching and a fail-fast gate.

The same work for you: CI that tells the truth, Premium, $2,600, or A server-side Swift API, from $1,500.

Work with me

Tell me the problem, and I will tell you which package fits, or that your job is larger than it looks. Email [email protected], or see the fixed-price packages.