Inbound Referrals

Track and process referrals coming into the clinic

Active Referrals

Patient	Referring Provider	Type	Status	Received	Actions

Drop-offs

Patient	Referring Provider	Days Since Contact	Attempts	Actions

Outbound Referrals

Patient	Destination Provider	Type	Status	Sent	Actions

All Referrals

Type	Patient	Provider	Referral Type	Status	Date

AI-Directed Workflow

How this prototype was scoped and built

Product leadership needed to see the scale of the referral workflow before a roadmap could be scoped. Rather than a static deck, I directed an AI coding agent through a structured brief so the team could click through a working artifact in days. Every step below was written or reviewed by me; the agent executed against it.

1. Brief the problem

What are you building?

Inbound and outbound referral tracking spanning intake, triage, scheduling, and closure, unifying e-referrals and physical fax/paper into one workflow.

What problem does it solve?

Eliminates the "black hole" where referrals get lost between disconnected channels, and gives staff clear tracking milestones so patients don't fall through the cracks.

2. Turn the brief into a PRD

Quantitative goals

Reduce referral drop-off rate by 50%
Cut time from receipt to patient contact by 75%
Raise loop-closure rate with referring providers to 90%

Personas

Front-desk staff, audiologists, and clinic administrators, each with distinct goals and pain points that shaped the triage and reporting views.

3. Break the PRD into an execution log

A Kanban-style issue list kept the agent scoped to one piece at a time, with explicit human checkpoints (HITL) on anything design-sensitive.

Scaffold mock data AFK

Populate realistic referrals, providers, and status vocabularies matching the architecture spec.

Build app shell & layout AFK

Header, sidebar navigation, and content regions matching the reference design.

Inbound / outbound / drop-off views AFK

Status progression tracking, filtering, and tab switching wired to mock data.

Review triage interaction model HITL

Design-sensitive step: I reviewed the modal, status stepper, and action set before it was signed off.

4. What I did after the agent's pass

Cleaned up layout, accessibility, and edge cases; restyled it to match the Sycle PRO design system used elsewhere in this case study; and turned the output into an executive-ready decision artifact rather than production code; this was built to align leadership on scope, not to ship.

Why this matters: a working, clickable prototype gave the CTO, COO, and VP a shared, concrete artifact to react to, replacing a slide deck with something they could click through in the room, which shortened the roadmap-alignment cycle from weeks to days.