Education Administration /Retrieval (RAG) live field guide · 8 min

Student Enquiry Concierge

Answers routine student and parent questions — deadlines, fees, entry requirements — from your own policies and FAQs, citing the document and version, and escalating anything it can't ground to a named team.

Open the live demo

theater/demos/edu-admin_student-enquiry-concierge.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

Reads the student's question, searches only your approved policies and handbooks, and returns a cited answer — or, when the answer isn't in your documents, says so and routes the enquiry to a named team for a person to pick up.

Input 01

The question + your documents

A student or parent's typed question, answered against an indexed library of your own policies, handbooks, fee schedules and FAQs — not a general internet search.

Agent 02

Retrieves, cites, scores

Searches the indexed corpus, drafts an answer grounded in the matching passages, and attaches the document name, section and version it drew from with a grounding-confidence score.

Output 03

A cited answer, or an honest escalation

A grounded answer with its sources quoted, or — when the question falls outside your documents — an explicit "not in our documents" that routes the enquiry to a named team for a person to answer and logs the gap.

Where it works well

It answers the same few hundred routine questions, 24/7, with the source quoted every time.

Best for providers with predictable enquiry peaks — open day, offer round, census date — where the same questions arrive in bursts.
Most valuable where a two-person admissions team handles high volumes, or international enquiries land outside Australian business hours.
The recaptured hours go back into the enquiries that genuinely need a person — complex eligibility, distressed students — not the repeat lookups.

The invisible cost in a small admissions or student-services team isn't the hard questions — it's the same handful of routine ones (census dates, fee deferral, entry scores, refund rules) asked hundreds of times, every one pulling someone off deeper work to retype an answer already written in a handbook.

Where it works badly

It is confidently wrong when the document behind the answer is stale, and the citation makes the wrong answer look authoritative.

Weak on questions that need a judgement call about an individual — a specific student's eligibility, a credit-transfer or RPL decision, an extenuating-circumstances case. It should escalate, not answer.
If your policies contradict each other across documents, it can surface either one with equal confidence — it retrieves, it doesn't adjudicate which policy wins.

The honest test

If you cannot say, today, that every document in the index is the current in-force version, this tool quotes your stale policy faster, not more safely.

Index last year's fee schedule or a superseded census-date procedure and it will quote it — section, version and all — in a clean, cited answer that reads exactly like a correct one. The citation is reassurance, not proof the document is current.

What it doesn't do — and shouldn't

It answers from your documents and escalates the rest. A person owns every non-routine reply.

WHAT IT DOES

Quotes the document name, section and version each answer came from

Says "not in our documents" rather than guessing when nothing matches

Routes the unanswerable enquiry to a named team and logs it as a gap

WHAT IT WON’T

Make an eligibility, admission, credit-transfer or RPL decision

Give an individual student personal advice on their case

Answer from the open internet or invent a deadline, fee or score

A wrong answer on a fee, a census date or an entry requirement can cost a student money or a place, and a provider's enrolment and records sit under ASQA's Standards for Registered Training Organisations and the Student Identifiers Act. The accountable staff member stays on every judgement call because the consequence lands on the student and the provider — not the tool.

What your data has to look like

A current, de-duplicated library of your own policies, versioned, with one in-force document per topic.

Typical readiness

across orgs we see, before the first job

Policies and handbooks as the indexed source

Needs shaping

A clear current version per document

Usual weak point

One authoritative answer per topic

Needs shaping

A named escalation owner per topic

Usual weak point

A captured log of what was answered

Usually ready

The real first job

The documents are usually the weak point — the live answer is a fee schedule on the website, a PDF in a shared drive and a staffer's memory, and they disagree. Getting to one current, versioned, de-duplicated source of truth per topic is usually the real first job — bigger and more valuable than the retrieval layer on top, because a clean corpus also fixes the humans giving answers, not just the tool.

Right fit if…

You field high volumes of routine, repeating enquiries at predictable peaks

Your policies and handbooks exist as documents you can point to and version

A small team is answering the same questions instead of doing deeper work

You have significant after-hours or international enquiry volume

Walk away if…

Most enquiries need a judgement call about an individual student's case

Your "current" policy is whichever document the asker happens to find first

No one owns keeping the fee schedule and procedures up to date

You want a tool that makes eligibility or admission decisions for you

Open questions

The worried-buyer questions, answered straight

It can surface a wrong answer if the document behind it is wrong or out of date — which is exactly why every answer quotes its source: the document name, section and version, plus a grounding-confidence score. When nothing in your library matches, it says “not in our documents” and routes the enquiry to a named team rather than guessing a fee or a census date. It never makes an eligibility or admission decision; a staff member owns those.

It answers only from the documents you index, so if your enrolment policy, fee schedule and census-date procedure contradict each other, it can quote either one with equal confidence — it retrieves, it doesn’t adjudicate which policy wins. Getting to one current, versioned, de-duplicated source of truth per topic is the first piece of work, and it pays off twice: cleaner answers from the tool, and clearer answers from your own staff.

No. It takes the few hundred routine lookups — census dates, fee deferral, refund rules, entry scores — that an admissions or student-services team answers over and over, so they spend their time on the enquiries that need a person: a specific student’s eligibility, a credit-transfer or RPL case, a distressed caller. Every non-routine question is escalated to them with context attached. The recaptured time goes back into student-facing work.

Current to the in-force version. VET fee rules and census dates change — a VET Student Loans debt is incurred the day after the census day, and a fee schedule or census-date procedure can be reissued between intakes. Index a superseded document and it quotes it confidently. The honest test: can you say today that every document in the index is the current in-force version?

A student’s enquiry, name and any identifier such as their USI are personal information under the Privacy Act and the Australian Privacy Principles, and a USI carries specific handling obligations under the Student Identifiers Act. Any deployment runs against your own systems and document store, not a shared pool — we scope where the data sits and who can see it as part of the build. The demo here runs entirely on fabricated documents and questions.

It declines to answer and escalates. Rather than inventing a plausible deadline or fee, it returns an explicit “not in our documents,” routes the enquiry to the named team that owns that topic — admissions, student services, fees — with the question attached, and logs the gap so a recurring one can be written up and added to the library.

What it takes to build

3–4 weeks · 4 phases

Reused from template~70%

Bespoke to this skin~30%

stack · Claude · private RAG · web widget

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 admissions or student-services team?

The honest place to start is a bite-sized first piece — index one policy set you can stand behind and see whether the cited answers hold. Tell us where the repeat questions pile up; we'll play it back, scope it, and show you what's possible.

Book a call How we work