Aged Care /Retrieval (RAG) live field guide · 8 min

Care Policy AskMe

Answers a carer's policy or procedure question at the point of care from your own approved library — quoting the section, naming the document and version, and saying so when the answer isn't there.

Open the live demo

theater/demos/aged-care_care-policy-askme.html · sandbox · read-only

Open

FIG. 1

The live demo, running on fabricated data. Open it to step through the full flow — every output is shown for a person to approve before anything happens.

How it would work

A carer asks a plain-language question, the tool retrieves the matching passage from your approved policy set and quotes it with document, version and section — then the carer reads the cited source and decides whether to act.

Input 01

A question + your approved docs

A carer's plain-language question at the point of care, and your indexed approved corpus — clinical governance policy, medication and care SOPs, the strengthened-Standards handbook.

Agent 02

Retrieves, grounds or refuses

Searches only the indexed approved passages, drafts the answer strictly from what it retrieved, and returns an explicit "Not in your approved documents" when nothing clears the relevance threshold.

Output 03

A cited answer, read by a person

A grounded answer with the document name, version and section quoted, plus a match signal — for the carer to read the cited source and decide whether to act, or to escalate a refusal to the right team.

Where it works well

It puts the cited line in front of the carer in seconds, instead of a 200-page PDF hunt.

Best for the carer or enrolled nurse on the floor, and the clinical governance lead who currently fields those questions by hand.
Every answer names the document, version and section it quoted — so the person reads the source, not a paraphrase.
Across a workforce asking the same procedural questions daily, the recaptured time goes back into resident care and governance judgement, not PDF archaeology.

The slow, invisible cost is the time floor staff lose hunting for an answer that is already written down — the retention rule, the medication SOP step, the restrictive-practices definition — buried in a policy on a shared drive that the carer at 2am cannot search.

Where it works badly

It will faithfully quote a superseded document — the failure is wrong-source retrieval, not hallucination.

Poor at questions needing judgement, not lookup — "is this reportable under SIRS?" depends on the facts of the case.
It can quote the eight SIRS reportable-incident categories and the Priority 1 (24-hour) vs Priority 2 (30-day) timeframes, but it cannot decide whether an event meets the threshold.
If the answer is not a sentence already written in an approved document, this gives you less than your own judgement does.

The honest test

If the answer to your staff's question is a sentence already written in an approved document, this fits; if it requires weighing facts against a definition, it does not.

If last year's medication SOP is still in the corpus, the tool cites it in good faith with an authoritative-looking version number. That is most dangerous on what changed recently — the strengthened Aged Care Quality Standards came into force on 1 November 2025 under the Aged Care Act 2024, so any pre-November handbook is now wrong about which Standard governs what.

What it doesn't do — and shouldn't

It surfaces the cited passage. A person applies it. That boundary is deliberate.

WHAT IT DOES

Quotes the document name, version and section it retrieved

Shows a match signal and the verbatim passage for the carer to read

Refuses explicitly and logs the gap when nothing in the corpus matches

WHAT IT WON’T

Decide whether an incident is reportable under SIRS

Reconcile two policy documents that disagree, or interpret a grey area

Instruct or act on the carer's behalf

A wrong call on a reportable incident or a medication procedure is a resident-safety and compliance matter under the Aged Care Act 2024 and the Serious Incident Response Scheme, answerable to the Aged Care Quality and Safety Commission. The accountable person — the clinical governance lead or document owner — stays on the decision because the consequence lands on them.

What your data has to look like

A single, current, correctly-versioned set of approved documents — one authoritative copy of each.

Typical readiness

across orgs we see, before the first job

One authoritative copy of each policy

Needs shaping

A real version marker and date per document

Usual weak point

Superseded versions retired from the corpus

Needs shaping

Machine-readable text, not image-only scans

Usual weak point

A re-index step tied to every policy update

Needs shaping

The real first job

Getting to one approved, current, machine-readable set — with a re-index step tied to every policy update — is usually the real first job. It is a matter of how documents are captured and governed, not of buying a tool, and it is typically bigger and more valuable than the AI layer that sits on top.

Right fit if…

You have a defined, approved policy set with one authoritative copy of each document

Floor staff keep asking the same procedural questions that have a documented answer

Your documents carry real version markers and dates, and superseded ones get retired

You can re-index whenever a policy is updated or a version bumped

Walk away if…

Your policies live as overlapping drafts and nobody is sure which copy is current

The questions you need answered require judgement — "is this reportable under SIRS?" — not lookup

Key documents are image-only scanned PDFs with no extracted text

You want a tool that decides what the policy should be or resolves conflicts between documents

Open questions

The worried-buyer questions, answered straight

It only answers from passages it actually retrieved from your approved corpus, and it shows the document name, version and section it quoted so the carer can read the source themselves. When nothing matches it returns “Not in your approved documents” rather than inferring — the demo does this with a leave-entitlements question, routing it to People & Culture instead of guessing. The carer is reading a citation, not taking an instruction; the person stays responsible for the action.

Then it will faithfully quote whichever version you indexed, including a superseded one. It cannot tell that a 2019 medication SOP has been replaced unless the current one is the document in the corpus, and an image-only scanned PDF can’t be indexed until its text is extracted. Getting the corpus to a single, current, correctly-versioned set is usually the real first job in aged care, and it is the part we help with before the AI layer earns its keep.

No. It surfaces what your documents already say; it does not decide what the policy should be, resolve a conflict between two documents, or interpret a grey area — under the Aged Care Act 2024 and SIRS those judgements carry regulatory consequence and stay with your clinical governance lead and document owners. What it recaptures is the time they and floor staff lose hunting through PDFs for a line that is already written down.

As current as you keep it — the tool answers from whatever you last indexed, so a re-index has to follow every policy update or version bump. This matters acutely right now: the strengthened Aged Care Quality Standards came into force on 1 November 2025 under the new Aged Care Act 2024, so any handbook predating that is out of date. We set the ingest so a new approved version replaces the old one rather than sitting alongside it.

The corpus and the questions stay inside your tenancy; the demo is built on private RAG over your own SharePoint, not a public chatbot. Policy documents generally are not resident health records, but questions can contain identifying detail, so access is scoped to staff and the exchange stays within your environment. We confirm the data-handling boundary with you before anything is indexed.

It refuses explicitly and logs the question as a gap for the document owner, and it can route the person to the right team — People & Culture for an Aged Care Award question, for instance. That refusal-and-log behaviour is deliberate: the unanswered questions become a list of where your policy set is thin, which is useful rather than hidden.

What it takes to build

3–4 weeks · 4 phases

Reused from template~70%

Bespoke to this skin~30%

stack · Claude · private RAG · SharePoint

What it would cost

Fixed scope, fixed price, fixed dates.

Bite-sized first piece

One contained change, low risk

Pilot build

Most builds land here

Embedded support

Scale on proof

Considering this for your org?

The honest place to start is a bite-sized first piece — usually getting your policy corpus to one current, correctly-versioned set. Tell us where it hurts; we'll play it back, scope it, and show you what's possible.

Book a call How we work