Thomas
All case studies

Case study — Behavioral-health EMR

Couve

An admissions and revenue platform built so a person knows what treatment costs before they say yes.

Role · product design & build, soloBuild · Django + HTMX, server-renderedTimeline · Feb–Jul 2026Status · live demo, synthetic data
Open the live demo →

One click to sign in · no credentials · all data synthetic

Couve dashboard — active census of 318, admissions pipeline by disposition with per-track status, and an estimated patient-responsibility pipeline of $896,256.
The whole operation, one screen. A live census of 318, every admission’s intake / financial / medical status rolled up by disposition, and an estimated patient-responsibility pipeline of $896,256.12 — computed, not typed in.

The problem

Four tools each knew half the truth about one patient.

Someone says the hardest yes of their life; three weeks later a $12,000 bill nobody warned them about lands. The coordinator never stood a chance — the answer lived in four tools, each of which shipped half of it:

The intake inbox
Caught every lead — then lost them to whoever remembered to follow up.
kept: capture everything
The payer portals
Held the real benefits — behind three logins, with three different answers.
kept: verify, don't guess
The cost calculator
Did the math — in a language only the biller spoke.
kept: the math itself
The census spreadsheet
Tracked everyone — four editors, no audit trail, one sort from chaos.
kept: one view of everyone

The thesis

Readiness is one answer, not five systems.

20,000+
real insurance plans in the pricing engine — CMS public-use data, so every estimate prices against an actual plan
5 → 1
tools and payer portals consolidated into one patient record — the readiness answer has one address

Every admission clears four tracks — demographics, financial, medical, disposition — rolled into one honest state on one patient record. The system’s job isn’t to store records; it’s to say whether this person is ready, and to refuse to pretend when they aren’t.

Admission Readiness

BLOCKED · VOB before admission: currently VOB Requested
One patient’s readiness, at a glance. Financial isn’t cleared — so the system blocks admission and says exactly why.You can’t skip the step that protects them.

See it run

You’re one click from the coordinator’s chair.

No credentials, no sandbox tour — the demo signs you straight in. Work the pipeline, open a lead, run an estimate, try to admit someone who isn’t cleared. The data is synthetic and resets itself; the behavior is real.

A walkthrough of the platform — census, pipeline, and a patient’s readiness resolving in place.
Couve pipeline board — active admissions as cards, each with demographics, financial, and medical status dots, grouped by disposition.
The pipeline board — every active admission as a card carrying its track status. An entire operation, scannable in seconds instead of a morning of reconciliation.

The decisions

Make the money knowable. Make the gates honest.

Cost before commitment

Nobody says yes to treatment blind.

A real verification of benefits becomes a plain-language, session-by-session estimate priced on the contracted rate. A frightened person can plan around a number; nobody plans around an EOB.

▶ Live — this is math, not a screenshot

1deductible
$800
2–7coinsurance
$200 ea
8+plan pays 100%
$0
Total patient responsibility, knowable on day zero$2,000
The estimator’s benefit walk, running in this page at a $800/day level of care. Drag the sliders — the patient’s answer updates the way the product computes it.
Gates that refuse

Some steps shouldn't be polite suggestions.

You cannot admit someone whose benefits are only “requested” — the system blocks it, in red, with the reason. Deliberate friction, placed exactly where skipping ahead becomes a surprise bill.

Built for 4 p.m. on a full census

Plain language, no billing-code fluency required.

Dispositions, human-readable timelines, contracted rates you can actually see. An EMR is measured by how fast one tired coordinator can find the truth.

White-label from day one

No org name is hardcoded anywhere.

Branding, facilities, payers, rates — all configuration. Any provider could stand this up without touching the code, because that discipline started at the first commit.

The machinery

One record, from first call to final payment.

The spine every screen hangs off. Because it’s one record, the dashboard’s revenue pipeline, the readiness gate, and the patient’s estimate can never disagree — they’re reading the same row.

Who did what, plainly

Solo build. Borrowed scars.

I designed and built all of it — schema, workflow, and interface. The judgment inside it isn’t mine alone: it comes from two years working inside a national behavioral-health provider, watching coordinators reconcile these systems by hand. Their workarounds are this product’s requirements.

What a production deployment would add is the load-bearing work a prototype gets to skip: a BAA-backed hosting posture, real payer integrations, and the compliance hardening that protects patients and staff — tracked in the repo as explicit debt, not hand-waved.

Where it landed

Kickoff to a live, seeded demo: five months, solo.

February: first commit. July: a deployed platform with a one-click demo, a 20,000-plan pricing engine, and a seeded operation exercising every readiness state — every claim on this page checkable by clicking around it.

▸ What I’m not claiming: production use, real patients, or measured outcomes — the data is synthetic and no PHI exists anywhere in the system. The claims here are the kind architecture can guarantee, and the demo will show you.

Next

Gab is the front door. Couve is the building.

Every lead here begins as a conversation. Gab, the intake assistant, hands off a lead that arrives demographics-complete — and the moment Couve verifies benefits, the out-of-pocket number can travel back into that conversation. One spine, from a scared 2 a.m. message to an admission with no surprises.