← All use cases / Care & NDIS
Synthetic data only
Care & labour-hire
Rostering meets care plans
An Australian care and labour-hire operator runs on email: aged-care facilities send lists of open shifts, a small office team retypes them into the care system, and the first agency to invite its staff wins the work. We built a working prototype of an AI rostering platform that reads those emails, gets invitations out in one click, keeps one live view of who is where, and drafts care plans, all without writing a single record into the incumbent system.
Explore the actual prototype
- Everything is synthetic: a fictional operator (“Fernhill Care”) with invented facilities, staff and shifts, no client data.
- A static mirror: 16 real prototype screens (office desktop views plus the nurse mobile app) frozen with sample data, so nothing is live and nothing can break.
- Nothing to sign up for: browse the office views and the nurse app exactly as they were built, and switch between them from the top bar.
Opens in a new tab · synthetic data only
The problem
- Facility emails list 50 to 100+ open shifts at a time. Each one is retyped by hand into the care system before anyone can be invited.
- It is a race. Other agencies receive the same email, and the first to put a nurse forward wins the shift. Slow intake is lost revenue, not just lost time.
- After-hours calls become a confused message from an answering service, and someone in the office sleeps next to their phone in case a 6am shift needs filling.
- A handful of staff forget to clock out every week. The office chases each one by hand, because facilities want proof of attendance and payroll depends on it.
- The incumbent care system’s API is read-only. Nothing external can create shifts in it, so any new tool has to work alongside the existing record without touching it.
open shifts can arrive in a single facility email, each of them retyped by hand before this build.
writes to the incumbent care record. The integration boundary is read-only by design: rosters can be read from the existing system, and nothing is ever written back.
What the prototype does
A shift email lands, the AI reads it
The office pastes or forwards a facility email. Claude extracts every shift, facility, date, time and role, and attaches a confidence score to each draft.
AI drafts
Nothing exists until a human approves it
The office reviews the drafts, corrects anything the parser was unsure about, and approves. Only then does a draft become a real shift on the board.
Human decides
Invitations go out in one click
Staff are matched by role and availability and invited by SMS. The database enforces first accepted wins, so a shift can never be double-filled, and the facility is notified once it is covered.
First accepted wins
One live view of who is where
Nurses accept shifts and clock in and out from their phone, with GPS on the clock event. The office home is exception-driven: not clocked in, missed clock-out, urgent unfilled, on shift now.
Live view
Care plans drafted from the referral pile
Upload the GP, OT and referral documents and the AI drafts a structured care plan, copying clinical facts verbatim rather than inventing them. Staff review it and upload it to the care record themselves.
Human signs off
How it runs in production
Hosted in Australia
The prototype runs end to end in Sydney regions, application and database. Production handling of real health information is scoped as its own signed item, never an afterthought.
The incumbent system stays
Clinical records and government claiming stay in the existing care system. This platform works alongside it, reads what it needs and writes nothing back, so the current record cannot break.
Humans sign off
The AI drafts shifts and care plans; it decides nothing. Every shift is approved by the office and every care plan is reviewed by a person before it goes anywhere.
What this page doesn’t claim
No invented ROI percentages, no anonymous five-star quotes. This is a working prototype built in days on de-identified data, live in front of the client as part of an open engagement. The read-only stance towards their care system is the agreed design boundary, and production handling of real health information is a separately scoped piece of work. When a build like this goes to production, the results section gets real numbers, and not before.
Related services
Racing other agencies to fill the same shifts?
Tell us where the hours go. If it looks like a fit, the next thing you see is a working prototype on your workflow, not a proposal.
