Service-Knowledge Concierge | Real Minds AI
Automotive /Retrieval (RAG) live field guide · 8 min

Service-Knowledge Concierge

Answers a technician or advisor's question from your own repair manuals, SOPs and right-to-repair data, with a citation for every answer — and says so when the answer isn't in the library rather than guessing.

concierge · sandbox · read-only
Open
FIG. 1

The live product. 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 workshop's own manuals, servicing policy and technical bulletins, answers a technician's question with the exact section it came from, and routes anything not in the library to a senior technician to confirm before it's acted on.

Input 01
The question + your documents

A technician's plain-English question — a fault code, a torque spec, a servicing-interval query — searched against your own indexed repair manuals, capped-price servicing policy and OEM technical bulletins.

Agent 02
Retrieves and cites

Searches only the indexed documents, returns the relevant procedure or specification, and attaches the exact manual, bulletin or section it drew the answer from — no answer without a source.

Output 03
A cited answer — or an honest gap

An answer with its citation shown for the technician to confirm against the vehicle in front of them, or an explicit "not in your documents" that escalates the question to a senior technician before anyone acts on it.

Where it works well

It answers the same procedural question every time, instantly, with the source attached.

  • Best where the answer genuinely lives in a document — a manual section, a servicing-policy line, a technical bulletin — and just needs finding fast.
  • Strongest for workshops with mixed experience levels, where the same procedural questions get asked of senior staff again and again.
  • The recaptured time goes back to the master tech's own jobs and to mentoring, not to fielding lookups all day.

The slow, invisible cost in a workshop is a junior technician stopping a job to hunt a torque spec or a servicing interval across three vendor portals — or interrupting the one master tech who knows it off the top of their head.

Where it works badly

It is confidently wrong when the document behind the answer is out of date — and the citation makes the stale answer look authoritative.

  • Weak on judgement calls — whether an aftermarket part affects a warranty position, whether a procedure suits this specific vehicle — where the answer needs reasoning, not retrieval. It should refuse, not infer.
  • If your document library is thin or scattered across personal folders, it has little to retrieve from and will refuse more than it answers.
The honest test

If you cannot say which version of each manual, policy and bulletin is in the library right now, this tool makes a wrong answer faster, not the right answer safer.

Index last year's servicing schedule or a superseded technical bulletin and it will return a clean, cited answer drawn from a spec that has since been revised. The citation makes it look more trustworthy than the document behind it actually is.

What it doesn't do — and shouldn't

It retrieves and cites. A technician confirms against the vehicle. That boundary is deliberate.

WHAT IT DOES
Returns the procedure or specification and the exact document section it came from
Refuses, and says so, when the answer isn't grounded in the indexed documents
Routes the unanswered question to a senior technician and logs it as a library gap
WHAT IT WON’T
Decide whether the procedure applies to the specific vehicle in the bay
Make a warranty or consumer-guarantee call
Authorise a repair or sign off the work

A wrong torque spec or servicing step can damage a vehicle or void a legitimate manufacturer-warranty position, and what a workshop tells a customer about warranty and repair choice is governed by the consumer guarantees in the Australian Consumer Law. A retrieved answer is a starting point; the licensed technician who confirms it against the vehicle carries the consequence, so they stay on the decision.

What your data has to look like

Your manuals, servicing policy and bulletins, indexed and known to be the current versions.

32%
Typical readiness
across orgs we see, before the first job
Repair manuals and procedures as searchable text
Usual weak point
A current capped-price / scheduled-servicing policy
Needs shaping
OEM technical service bulletins, version-dated
Needs shaping
Right-to-repair service and repair data you've purchased
Usual weak point
A clear owner for what's in the library
Needs shaping
The real first job

Most workshops hold their knowledge across a few senior technicians' heads, vendor portals and ad-hoc folders, with no single current document set. Getting the manuals, the servicing policy and the bulletins into one indexed library — and keeping the versions current — is usually the real first job, larger and more valuable than the retrieval layer on top. Once the library is clean and current, every answer after that is fast and grounded by default.

Right fit if…
Your procedural knowledge already lives in manuals, policies and bulletins — it just isn't fast to find
You have a mix of experience levels and senior techs spend real time answering the same lookups
You can name an owner who keeps the document library current and retires superseded versions
You've purchased right-to-repair data and want a practical way to use it without everyone navigating the portal
Walk away if…
Your knowledge is mostly in people's heads, not in documents the tool can index
Nobody owns keeping the servicing policy and bulletins current
You want it to make warranty or consumer-guarantee calls rather than route them to a person
Most of your questions are vehicle-specific judgement calls, not document lookups
Open questions

The worried-buyer questions, answered straight

It can only return what’s in your indexed documents, and it attaches the exact manual, policy or bulletin section every answer came from — there is no answer without a source to check. When the question isn’t grounded in your documents, it says so and routes it to a senior technician rather than guessing. The licensed technician confirms the answer against the vehicle in front of them before acting; the tool surfaces the source, the person stands behind the work.
It retrieves from text it can index, so scanned-image PDFs with no text layer and documents trapped in vendor portals give it nothing to search. Getting your repair manuals, capped-price servicing policy and technical bulletins into one indexed, current library is usually the first piece of work — and the piece that pays off on every answer after, because the tool is only ever as good as the documents behind it.
No. It removes the lookups — the torque spec, the servicing interval, the bulletin reference a junior tech would otherwise interrupt a senior tech to get — so the master technician spends their time on diagnosis, judgement and mentoring rather than fielding the same questions all day. The recaptured time goes back into the bay and into the team, not out of the headcount.
Current to the versions in force today. Servicing schedules get revised and OEM technical bulletins get superseded, and the tool can’t tell a current document from last year’s — it will cite a stale spec just as confidently as a current one. Someone has to own adding new documents and retiring the superseded ones; without that, the citation makes an out-of-date answer look authoritative.
The document library is indexed inside your own tenancy, not a shared cloud pool — your manuals, servicing policy, bulletins and any right-to-repair data you’ve purchased stay within your systems, and we scope where they sit and who can query them as part of the build. The demo here runs entirely on fabricated workshop documents.
No, and it’s designed not to try. A warranty position depends on the specific vehicle and on the consumer guarantees in the Australian Consumer Law, which a manufacturer can’t simply exclude because an independent repairer or aftermarket part was used — the ACCC has taken enforcement action over misleading warranty claims. That’s a judgement call, not a document lookup, so the tool refuses and escalates it to a person who can give the customer a correct answer.
What it takes to build
3–5 weeks · 4 phases
Reused from template~70%
Bespoke to this skin~30%
stack · Claude · vector store · retrieval · review UI
What it would cost

Fixed scope, fixed price, fixed dates.

01
Bite-sized first piece
One contained change, low risk
02
Pilot build
Most builds land here
03
Embedded support
Scale on proof

Considering this for your workshop?

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

More in Automotive
Warranty-Claim Drafter
View →
Review-Response Drafter
View →
Booking & Phone Agent
View →
Missed-Call Rescue & Booking Agent
View →
How We Work Proof Talk to us
How We Work Proof Talk to us
Ask us anything