iSleep + uNITE Sleep / sleep medicine / regulated healthcare / one operator
A HIPAA-grade platform, the analytics beside it, and the cloud under both, for a sleep medicine company.
Boring on purpose.
11 wks
first commit to production
5,955
test cases in the suite
201
API routes on the platform
1
operator on the whole thing
The platform, on screen.
Real production surfaces from the member platform and the admin console. Names and identifiers are covered, and the data on screen is synthetic QA.




Drift: three portals, one compliance engine.
Drift is the platform those screens belong to. Members onboard in four steps, watch their own nights against the insurance goal, and reorder supplies through a pipeline that clears eligibility before anything ships. Providers get the same data with a billing layer on top. All of it from the QA run, on my own test record.






Moonbase: the design department, as software.
The print program below outgrew shared folders, so I built Moonbase: every brand asset organized and versioned, a design request flow any staffer can drive, and an AI pipeline that drafts new collateral from approved templates and outputs print-grade PDF. It syncs both ways with the company’s project tracker.



A homepage and a landing system, designed in-house.
The same year redesigned the public storefront: a new homepage and a landing page for every front door into care. Sleep apnea, CPAP therapy, insomnia, sleep hygiene. These are the design files; the AI page system further down carried them to production.






Two brands, one designer.
The same year produced the creative direction for both brands: voice documentation, the uNITE Sleep Institute launch collateral, provider flyers, paid social systems, and the print program below, from brochures to a rodeo program ad to the holiday coffee label. Every piece shipped.














A conference booth, designed at fabrication scale.
When the company booked a trade show, the booth came through the same one-person studio: a back wall with a built-in screen, a before-and-after side panel, the header banner, and the counter wrap, all drawn at print dimensions for the fabricator. The booth TV ran the ad program below.




Seven ad concepts, script to publish.
Written, generated, edited, and published as a running ad program. Still in rotation on the channel, and the reel the booth screen carried. Tap to play.
The one-operator model, on hard mode.
A regulated healthcare company with members on one side, clinicians on the other, and insurance rails in between. The work started in brand and creative direction, and the scope grew the way it does when one operator can hold the whole map: storefront, martech, analytics, the member platform, the cloud estate.
The platform went from first commit to production in eleven weeks.
That speed is only safe because the floor is overbuilt. 5,955 test cases, field-level encryption on the identifying columns, an access log on every administrative read. Compliance work rewards the operator who finds that kind of thing fun. I do, a little bit.
One map, five zones, real counts.
The actual topology of the compliance platform, drawn from the repo instead of a slide. Vendors appear as categories because naming them would fingerprint the client. Click through the zones; every number in here is real.
Three portals, one data layer.
Admin console, clinician portal, member portal. All three render off the same API surface, so a compliance rule changes in one place and holds in three.
all three call
over
- Portal experiences
- 3
- Staff role types
- 5
- API route files
- 201
- Authorization
- Role-scoped at the app layer
// counts verified against the production repo before publishing · vendor names withheld on purpose
FIELD NOTE / THE COMPLIANCE FLOOR
If the audit log is the most interesting thing in your architecture, something has gone wrong. Mine is the dullest table in the database. It gets written to more than any other.
18 ACTION TYPES / TWO SINKS / EVERY ADMINISTRATIVE READ
Fifteen jobs run the clock.
Eligibility re-verifies at 2am. Compliance metrics recalculate at 3. The queue dispatcher never gets to stop. This is the platform’s real schedule, plotted on the wheel it actually runs.
- Call queue dispatch + health monitorevery 5 min
- Voice session reconcilerevery 10 min
- Fulfillment order submissionevery 2 h
- Fulfillment status syncevery 4 h
- Payment outreachevery 4 h
- Device telemetry syncevery 6 h
- Insurance eligibility re-verification02:00
- Report artifact cleanup02:00
- Compliance metrics recalculation03:00
- Catalog sync04:00
- Compliance + data-retention check08:00
- Referral pipeline monitor08:30
- New-member onboarding outreach09:00
- AI SMS outreach10:00
// real schedules from the production deploy config · job names at category level
// schedules read from the production deploy config · static render, no animation to wait for
Warehouse first, dashboards second.
The first console went from empty repo to five live pages in twenty commits on a single afternoon, wired straight to commerce and ads data. It proved the appetite. The enterprise rebuild followed in thirty days: a real warehouse, a Python API, and a live order feed over websockets.
17
data sources
commerce, ads, email, search, support
1
ingestion layer
webhooks + scheduled backfills
38
warehouse models
staging to marts, tested
136
API endpoints
25 routers, one websocket feed
20+
dashboards
exec, product, cohort, live
// v3 console: 20 commits, one afternoon · v4 rebuild: 79 commits, 30 days · source categories only
Platform plus analytics was the spine. The same engagement carried the brand side.
That pairing is the point of one operator. The person wiring the access log also art-directed the print files, and neither waited on a handoff.
Moonbase, by the numbers
The brand hub above: AI-assisted templates, print-grade PDF output (PDF/X-1a, CMYK, 300 DPI), and routing straight to the print vendor. 86 commits over seven weeks.
Storefront page system
An AI page-generation pipeline with 158 reusable section templates behind 6 production landing pages on the commerce storefront. The web layer above shipped through it.
Cloud estate, documented
An inherited estate of 90+ cloud functions and 47 storage buckets, inventoried function by function: 68 runbook and architecture docs covering 23 systems and 8 partner data pipelines.
SEO + editorial
Fourteen long-form articles for the organic surface, plus a Python SEO intelligence layer on Postgres + pgvector for interlinking and keyword-gap work.
5,955
test cases
330+
API endpoints, two platforms
116
database tables
15
scheduled jobs
17
voice-agent tools
38
warehouse models
90+
cloud functions documented
68
runbook + architecture docs
Regulated work punishes handoffs.
Every boundary between the person who knows the compliance rule and the person who writes the query is a place where member data can leak. Most companies staff those as two people and spend real energy managing the seam. The consolidation that makes a one-operator stack fast in DTC is the same thing that makes it safe in healthcare: the whole pipeline lives in one head, so the floor holds everywhere.
The client’s name stays off this page. The receipts still carry it: 5,955 tests, the schedule the platform keeps at 2am, and an access-log pattern I’d build again tomorrow.
A member platform in production behind three portals, a warehouse-backed analytics suite, a two-brand creative program with a sub-brand launch, and a cloud estate with real documentation. One operator on the whole thing.
The Tracking Gap
Q2 2024Browser-only tracking was missing half the story.
NDA client
The Agent Council
Q1 2026Five named agent personas, each modeled on a real exec, all orchestrated through a single Claude Code skill.
NDA client
The Rooted Life
Q1 2026Nutritional therapy brand, course, and clinical platform. Solo build.
The Rooted Life
Regulated work needs a steady pair of hands.
I take this on as a monthly retainer: the platform, the compliance floor, the analytics, and the brand surface, handled together. DTC and healthcare companies in the $2-10M range, mostly. It’s just me on the file.