The problem: dispatching trust across 70,000 square miles
The business idea was simple: patient, in-home tech help for people who would rather have a polite person at the kitchen table than a support line in another time zone. The operating problem behind it was not simple at all. It meant dispatching vetted contractors across a service territory of nearly 70,000 square miles, pricing every drive honestly, protecting the customer from the contractor and the contractor from the customer, and doing all of it on margins that survive a rural service call.
No off-the-shelf product solves that. Hosted scheduling tools manage one person's calendar; this needed a fleet, with travel time, skills, background-check gates, and per-distance pricing decided at the moment of booking. So instead of stitching the company together from a dozen subscriptions, I architected one platform that runs the entire business end to end, and led the build from systems design through delivery.
The architecture, at a glance
The design is deliberately simple: every surface a customer or technician touches talks to one engine, so a business rule changed once is changed everywhere. Simple systems are cheap to run, fast to change, and easy to reason about when something goes wrong.
What one platform carries
Everything below ships from one codebase, behind one API, under one set of engineering standards.
A purpose-built engine models every technician's day as precise intervals. Holds, bookings, and dispatches are the same kind of record, so on-demand service is configuration, not a rewrite.
Returning customers get the technician who helped them last. Skills profiles keep the wrong person from being sent to fail. Coverage is computed from real committed drive ranges.
Memberships, one-time visits, in-field charges, and tips on one set of rails, with versioned terms: a price change never rewrites what an existing member agreed to. Card data never touches our servers.
The phone system's intelligence lives in our own code on a usage-priced carrier: one greeting for every line, masked relay calls, and local numbers chosen per visitor.
Self-hosted remote desktop for computers, plus a purpose-built iPhone screen-sharing path that mainstream remote-support products simply do not offer. One link does both, and the customer's device decides which.
A Nerd can take a card at the kitchen table: a keyed field that appears on the sign-off screen when there is no card on file, or a terminal opened against that one visit and filed as a visit balance, so a day never double-counts. Contactless acceptance runs on the Nerd's own iPhone, no extra hardware to carry.
Coverage is computed live per ZIP code against every market anchor, so there is no lookup table to fall out of date. A ZIP nobody can reach is recorded as demand turned away rather than quietly served remote.
A customer app and a field app for Nerds, each built twice: native Swift on iPhone, native Kotlin on Android, with no web shell anywhere. Deliberately mirrored so a feature lands on both platforms in step. Widgets, app intents, push, and native payment sheets included.
Thirty permissioned modules for books, payouts, coverage, hiring, and messaging, with access enforced server-side per module so no role quietly grows into another.
Each technician's cards are composed from live data: right phone number, right price, their own referral code, proofed at press geometry and drop-shipped to their door.
A branded merchandise storefront, public-facing and internal, fulfilled print-on-demand with tax, tracking, and shipping notifications handled. Zero inventory, zero risk, and the brand on every box.
Decisions that protect revenue and trust
These are the calls a buyer should ask any architect about, because they are invisible until the day they are everything.
The conflict check and the write that claims a slot happen as one atomic transaction. Dedicated tests race two customers for the same slot to prove it.
A job blocks the drive on both sides of it, inside the technician's stated hours. Nobody's day gets silently consumed by double-counted drive time.
A visit that runs long never silently rewrites the agreed price. Overage goes to a human, never an auto-bill, with an append-only audit trail behind every visit.
AI triage estimates realistic durations from the customer's own words, but it is advisory: if it is down, the booking completes. Calendar sync can subtract availability, never add it.
Contractors never see a customer's phone number or email. Not hidden by the interface: structurally absent from every payload they can reach. Webhooks are signature-verified; staff sign-in is SSO with server-side checks.
Every deploy audits its own pages and claims against their source of truth and ends by polling the live health endpoint. If any check fails, nothing ships. Backups are integrity-checked nightly into three failure domains.
Owned where it matters, rented where it should be
The build-versus-buy line was drawn deliberately: own everything that is the business, rent only commodity utilities priced by use, and never let anything rented hold the customer relationship or the data.
| Capability | How it is delivered |
|---|---|
| Scheduling & dispatch | Owned engine, built for a fleet; hosted tools schedule a person |
| Website | Owned static site, generated from one source of truth; no CMS to hack |
| Phone system | Owned call logic on a usage-priced carrier; no per-seat licenses |
| Remote support | Self-hosted; no per-technician licenses |
| Payments & screening | Deliberately rented: regulated money movement and compliance stay with specialists, priced per transaction |
| Marketing collateral | Owned variable-data renderer; trade printing per order, no minimums, no inventory |
| Branded gear | White-label storefront on per-item print-on-demand fulfillment; zero inventory risk |
The result is a stack with no per-seat software anywhere in it: the tenth technician and the hundredth cost the same software dollars, which is zero, and the marginal cost of serving the next customer or the next market is close to nothing. Twenty markets were mapped, priced, and wired in before launch and thirteen are open today, so opening the next city is a switch in the back office rather than an engineering project.
The asset test: over 180,000 lines of first-party application code behind 429 endpoints, another 111,000 lines across the four native apps, and a suite of more than 3,000 automated tests, all built since June 2026. It has been security-reviewed, it takes live payments, and it is backed up in three places. That is the difference between a stack of subscriptions and a company that owns its own machinery.
What this means for your project
OkieNerd is proof of full-scope delivery: brand identity, product design, platform architecture, four native apps, payment rails, telephony, security posture, and the operational tooling to run it, all under one accountable architecture. That is what I bring to an engagement: I design the system, set the engineering standard, and stay hands-on with the team building it, so the vision and the execution never drift apart. If your business needs software shaped to how it actually operates, connected to everything you already run, and owned rather than rented, this is what that looks like carried all the way to production.