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

Student Query Concierge

Answers routine student questions on deadlines, enrolment, fees and policy from your own handbook 24/7 — quoting the section and date, and escalating to a person when the answer isn't there.

Open the live demo

theater/demos/higher-ed_student-query-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

Searches only your approved student documents, drafts a cited answer to a student's question, and refuses — escalating to a staff member — when the answer isn't in the corpus.

Input 01

The question + your documents

A student's plain-language question through Teams, the web widget or the student portal, answered against your indexed corpus — enrolment and census procedure, fee schedule, special-consideration guidelines, academic progress policy, the student handbook.

Agent 02

Retrieves, gates, drafts

Searches the indexed chunks by similarity, scores the top match against a confidence threshold, and drafts an answer grounded only in the retrieved passage — with the document, section and date cited.

Output 03

A cited answer, or an honest refusal

A grounded answer carrying its source quote and a Helpful / Send-to-student control, or — when nothing clears the threshold — a refusal that escalates to a staff member. A person confirms or sends before the student relies on it.

Where it works well

It answers the same predictable questions every time, in full, with the section and date cited.

Best where a known set of questions dominates the queue and the answer already exists in an approved document.
At peak — orientation, census week, results release — it absorbs the volume a small team can't, around the clock.
The recaptured hours go back into the cases that need a person — complaints, special consideration, at-risk students — not the FAQ.

The invisible cost is the repetition tax. At intake, census and exam time the same handful of questions — when is the census date, what happens if I drop a unit after it, how do I apply for special consideration — flood the desk, and each one is a small interruption that lands on student services.

Where it works badly

It is confidently wrong when your documents contradict each other or a superseded version is still indexed.

Weak on judgement questions — whether a student's circumstances meet the special-consideration threshold is a decision the document can't make for them.
If your policy content lives in scattered shared-drive versions with no single owner, the tool surfaces that mess fast and confidently.

The honest test

Pick five questions your students actually ask and check whether each has one current, approved document that answers it in writing. If most don't, the document work comes first — retrieval can't manufacture an answer that isn't written down.

Index last year's enrolment procedure alongside this year's, or two handbooks that disagree on a deadline, and it will retrieve and cite one of them with the same confidence it shows for a clean answer. It has no way to know which is the current, authoritative version.

What it doesn't do — and shouldn't

It drafts a cited answer. A person sends it, and a person answers what it refuses.

WHAT IT DOES

Shows the document, section and date it relied on, plus the verbatim quote

Flags its retrieval confidence so a weak match is visible

Refuses and escalates when nothing in the corpus clears the threshold

WHAT IT WON’T

Decide a special-consideration, appeal or fee-remission outcome

Send anything to the student without a staff member's confirmation

Answer from anything outside your approved documents

A wrong answer about a census date, fee liability or an academic-integrity process has real consequences for a student's money and standing — and the provider's obligations to maintain accurate student information sit under the Higher Education Standards Framework (Threshold Standards) 2021, which TEQSA regulates. The accountable staff member stays on the answer because the consequence lands on the student, not the tool.

What your data has to look like

One current, approved version of each student-facing document, with a visible version and date.

Typical readiness

across orgs we see, before the first job

Approved student documents as the corpus

Needs shaping

One current version of each, clearly owned

Usual weak point

A version and effective date on every document

Usual weak point

Re-indexing built into document approval

Needs shaping

A channel for students to reach it

Usually ready

The real first job

The documents usually exist; the discipline around them usually doesn't — three versions of the enrolment procedure, a handbook that hasn't tracked the latest fee change, no single owner for "what's current". Establishing document control so there is exactly one live version of each is usually the real first job — larger and more valuable than the retrieval layer on top. Once the corpus is clean and current, every answer after that is fast and right by default.

Right fit if…

A known set of repetitive questions dominates your intake, census and exam-period queue

You have approved student documents — handbook, fee schedule, enrolment and census procedure

You can point to one current, in-force version of each document today

Student services is the bottleneck on lookup questions a document already answers

Walk away if…

Your policy and handbook content lives in scattered, conflicting shared-drive versions

Most inbound enquiries are judgement calls — special consideration, appeals, complaints

No one owns "what's current", so superseded versions stay live

You need a tool that decides outcomes rather than quotes the policy

Open questions

The worried-buyer questions, answered straight

It can — which is exactly why nothing goes to a student on its say-so. It only answers from passages that clear a confidence threshold, and it shows the document, section and date it relied on plus the verbatim quote, so a staff member checks the source, not just the summary. When nothing in your approved documents matches (the demo’s parking-permit question), it refuses and escalates to a person rather than guessing. The safeguard is that the answer is never separated from its source, and a person confirms anything consequential.

Not well, and it will make the mess visible fast. The tool answers from whatever you index, so if two documents disagree it can cite the wrong one or surface a superseded version with full confidence. Sorting out which document is current and authoritative — one version of each, clearly owned — is usually the real first job in an Australian higher education provider, and the piece that pays off across every answer after.

No. It answers the high-volume “where is it written” questions — census dates, fee calculation, how to apply for special consideration — so your student services team isn’t the lookup desk. That recaptures their time for the work that needs a person: an at-risk student, a complaint, a special-consideration decision, an appeal. The tool quotes the document; it can’t decide an outcome or judge a student’s circumstances, and it routes anything it can’t answer to a person.

As current as the version you’ve indexed — the tool is only ever as right as that. It shows the version and effective date on every citation, so a stale answer is visible rather than hidden, but it has no way to know a document is out of date. Re-indexing when a policy, fee schedule or enrolment procedure is approved, and retiring superseded versions, has to be part of your document-control process. The honest test: do you know, today, which version of the fee schedule this is drafted against?

The corpus is your approved documents only, and a production build runs against a private retrieval index you control — not a public model’s training data. Student questions are personal information handled under the privacy law that applies to your provider — the Privacy Act 1988 and the Australian Privacy Principles for private providers and ANU, or the relevant state or territory privacy law for most public universities — and providers must keep student records securely under the Higher Education Standards Framework’s information-management standards that TEQSA oversees. We scope data residency and access with you up front. The demo here runs entirely on fabricated data for NorthLink AI, a fictional provider.

What it takes to build

3–4 weeks · 4 phases

Reused from template~70%

Bespoke to this skin~30%

stack · Claude · private RAG · Teams/web

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 institution?

The honest place to start is a bite-sized first piece — one contained change, low risk. Tell us where the queue hurts; we'll play it back, scope it, and show you what's possible.

Book a call How we work