What devices does Loomwork pull data from to pre-populate Epic?

Barcode scans from Zebra handheld devices, vitals from Welch Allyn monitors, and infusion-pump events from Baxter Sigma Spectrum pumps. Loomwork writes the resulting documentation back into Epic using the standard FHIR write APIs. The architecture is intentionally read-many, write-narrow: Loomwork ingests broadly across plant devices but only writes a curated, draftable subset back into the EHR, which is what allowed the post-rewrite product to back off cleanly when a nurse marks a patient complex.

What did Patricia's email change about Loomwork's product?

Patricia, a nurse at the pilot hospital, sent a 1,200-word email at 2am detailing how Loomwork's pre-populated documentation worked well for the 70% of stable med-surg patients but actively harmed the 30% of high-acuity patients (rapid deterioration, post-surgical complications, transfers). Loomwork rebuilt the product in four months around a 'complex patient' mode that does less automation, makes drafts obviously overridable, and never insists. Documentation savings dropped slightly on aggregate but nurse NPS climbed from 31 to 78. Patricia is now a paid clinical advisor.

How does Loomwork stay HIPAA-compliant in its observability stack?

All third-party services that touch PHI sit under signed HIPAA BAAs — AWS RDS for the Postgres database, AWS Kinesis for the event ingestion pipeline, and Datadog for production observability (Datadog supports a HIPAA-eligible plan). The Next.js clinician dashboard runs on Vercel but only handles tokens and identifiers, never raw PHI; PHI lives behind a separate API tier. Linear is used for issue tracking but no PHI ever appears in tickets, by policy and by automated linting on issue creation.

Healthtech founder story: how Loomwork rebuilt after one nurse's email

IChapter 1 of 5

The Problem

Clinical workflow software is famously broken. Nurses at most US hospitals spend 35-45% of their shift on documentation in an EHR — Epic, Cerner, Meditech — that was designed primarily for billing, not for clinical work. Every health-system CIO knows this. Every nurse knows this. And yet the dominant EHRs sit at the center of contracts so large and integrations so deep that ripping them out is unthinkable.

The opportunity, if there was one, lay in the layer above the EHR. A workflow product that lived between the nurse and the EHR, automated the documentation drudge, and gave the nurse back fifteen minutes a shift. Multiply that across a 600-bed hospital and you save roughly $4-6M a year in nursing labor.

I'd been a clinical informaticist for seven years. I'd watched startups try to build this layer and fail. Most failed because they were built by software people who'd never held a patient's hand at 3am. The product would pass demo with a CIO and die on contact with the actual unit.

Loomwork's bet was that a clinically-led founding team — myself, a hospital pharmacist named David, and an EHR-integration engineer named Ren — could ship the layer that previous startups had missed. I'd read a piece on representation learning over EHR data the same week we incorporated, which mostly confirmed our suspicion: the data was rich enough; the workflow layer was the problem.

The Breakthrough

What we saw on those shifts changed the company. We rebuilt the product over the following four months around a single principle: high-acuity moments are not exceptions, they are the work. The pre-populated documentation had to be obviously visible-as-draft, easily overridable, and never insistent. The app needed an explicit "complex patient" mode that did less automation, not more. Nurses needed a way to flag a patient as complex and have the system back off.

We shipped the rewrite in February 2024. Documentation time savings went from 18% (averaged) to 24% on stable patients and 9% on complex patients. The total reduction was smaller in aggregate. Nurse NPS at the pilot hospital went from 31 to 78.

The CIO renewed the pilot for the entire hospital in May 2024 at $640k ARR. As of April 2026 we have 14 hospital customers, $5.8M ARR, and a team of 21. Patricia is a paid clinical advisor. She has a quote on our homepage.

The broader lesson — that aggregate adoption metrics had hidden a real product problem from us — has changed how we look at every single metric we measure.

§What the second version of the product proved

The nine months after we shipped Loomwork v2 — the rebuild that came out of Nurse Patricia's email — were the most validating stretch of the company. Nurse-reported satisfaction (a metric we now collect on every shift via a thirty-second post-shift survey) moved from 4.1/10 to 8.6/10. Average documentation time per patient dropped from 14 minutes to 5.3. Our two anchor hospital customers — a 410-bed facility in Pittsburgh and a 280-bed in Phoenix — both expanded from a single med-surg unit to system-wide deployment within six months of v2 launching.

We made two hires that mattered. Patricia herself joined as Director of Clinical Workflow in late 2025 — she's now the person every product spec is reviewed by before it ships. Her existence prevents us from making the v1 mistake again. The second hire was an integration engineer named Bao, who'd worked on Epic Hyperspace at his previous role and rebuilt our HL7 FHIR layer to interoperate with Epic, Cerner Oracle Health, MEDITECH, and Allscripts without per-customer custom work. ARR crossed $4.8M in March 2026 with seventeen people. Our HIPAA + HITRUST CSF posture passed our second audit cycle in February without findings — the AWS RDS, AWS Kinesis, Datadog HIPAA-eligible, and Auth0 B2B stack we standardized on has held up through two regulatory reviews and a Joint Commission survey at the Pittsburgh hospital.

The Lessons

1
Aggregate metrics in healthtech are dangerous.
They average across patient acuities, shifts, and units. The patients who matter most are usually the minority case your dashboard hides.
2
The 2am email from the front-line user is the most valuable signal you will receive.
Read it twice. Do not respond defensively. Get on a plane.
3
Clinical software has to back off, not push.
Default behavior should never insist when a clinician overrides. The opposite default is how clinicians come to hate every EHR.
4
Shadow shifts are not optional.
Every product decision about clinical workflow should be made by someone who has spent at least one full overnight shift on the floor. If your team can't, your product won't.
5
Make the user a paid advisor.
Patricia is a clinical advisor. The arrangement is honest, transparent, and fairly compensated. The signal it sends to other front-line nurses is irreplaceable.

What this looked like in practice.

The thing nobody tells you about *building healthtech for nurses* is that the buyer is the Chief Nursing Officer and the user is the floor nurse, and they almost never agree about what "good" looks like. We learned to ship the CNO a weekly utilization report and the floor nurse a weekly time-saved report, and both grew to defend us internally.

The thing nobody tells you about *clinical workflow software* is that documentation is not the bottleneck — context-switching is. A nurse who clicks between four screens to log one observation is the same nurse who burns out in eighteen months. We track screen transitions per documented event as our north-star UX metric.

The thing nobody tells you about *HIPAA and HITRUST* is that the certification is the easy part — the operational discipline of staying compliant under feature pressure is the hard part. We have a written rule: every PR that touches PHI handling requires sign-off from two engineers, one of whom is on the security team. That rule has saved us from at least four near-incidents.

The thing nobody tells you about *integrating with Epic* is that there is no shortcut. Epic Hyperspace plus the App Orchard review process is a six-to-nine month effort that no amount of cleverness compresses. We baked that timeline into our enterprise sales cycle from day one.

The thing nobody tells you about *hiring the customer who emailed you* is that they will be the highest-leverage product hire you ever make. Patricia's $190k salary has paid for itself in six months of avoided product mistakes.

We rebuilt our entire clinical product after one nurse's email

The Problem

The Journey

§The first hospital pilot

The Struggles

§What the metrics had hidden

The Breakthrough

§What the second version of the product proved

The Lessons

Frequently asked

Helena Vasquez

Responses

Join the conversation