Overview
AR Car & Bike Advisor turns your phone camera into a research companion for buying a vehicle. Point it at a car or bike in a showroom, a car park or on the street and capture a frame; a vision model makes its best guess at the make, model and likely variant, and an augmented-reality spec panel is drawn over the frozen frame with the typical engine, power, mileage, seating and fuel options, an approximate price range, and — always — a confidence figure.
From the identified vehicle you can go four ways: work out an estimated cost of ownership over the years you plan to keep it, put two or three vehicles side by side and score them against the priorities you actually care about, prepare for an honest conversation with the dealer, or find dealers near you. The whole scan-to-shortlist path works with no account at all; a free login saves your garage, and a flat monthly premium removes the free-scan limit.
The product is deliberately built to under-state. It is a starting point for your own research: it gives ranges rather than precise numbers it cannot know, states its assumptions next to its figures, reports how confident it is, and tells you to confirm everything with the dealer or manufacturer before you act.
The honesty framing
This is a money product on the buyer's side, so over-stating would do real damage. Four rules shape every screen.
Every figure is an estimate
Prices, mileage, running costs, resale — all of it is approximate and given as a range, never as a precise number the model cannot know. The disclaimer the AI returns — “AI estimate for research only. Confirm specs, prices and figures with the dealer or manufacturer” — is shown prominently on every result, never as fine print.
Identification is a best guess
Recognising a vehicle from one photo is genuinely hard — trims look alike, facelifts differ by a bumper, angles and light mislead. So the scan always shows its confidence number, says plainly that the specs are typical for that model rather than measured from the exact unit photographed, and offers a “not right? correct it” control so you can fix the make and model before anything else is calculated on top of it.
Assumptions travel with the numbers
A cost-of-ownership total means nothing without the assumptions behind it. The calculator shows its assumptions card first, above the breakdown, so you can see the fuel price, the running distance, the service intervals and the depreciation view it used — and disagree with them. No single hard number is ever presented on its own.
No brand disparagement
Comparisons are made only against the priorities you state, with the trade-offs shown in both directions. Every vehicle in a comparison — including the one the AI picks — carries a trade-off, and the recommendation card always ends with who should pick the other one instead. The product does not run down any brand, dealer or manufacturer.
What it does
1. Scan & identify
Open the camera, point it at a car or bike and capture — or upload a photo you already have. The frame is sent to a vision model which returns its best guess at the make, model, likely variant, body type and year range, with a confidence figure. If it cannot identify the vehicle it says so and offers manual entry rather than guessing.
2. AR spec panel
The identification is drawn as a glass spec panel pinned over the frozen camera frame: make, model and variant guess, a confidence chip, the typical engine, power, mileage, seating and fuel options, and an approximate price range. A glanceable read on the vehicle in front of you, with the full spec sheet one tap away.
3. Cost of ownership
The sticker price is only the start. Tell it how many years you plan to keep the vehicle, how far you drive a year, the fuel type and your city tier, and it estimates the cost heads — purchase, fuel or charging, insurance, servicing, tyres, depreciation — with a note on each, a total, a per-kilometre figure and a plain-language narrative. The assumptions card comes first, and a doughnut chart shows where the money actually goes.
4. Compare your shortlist
Add two or three vehicles and pick the priorities that matter to you — budget, mileage, space, safety, performance, resale, maintenance. You get a matrix of criteria scored out of ten per vehicle, the genuine strengths and trade-offs of each, and a balanced recommendation that names who should pick the other option instead. Scored against your priorities, not a generic verdict.
5. Fair-deal coaching
Type the vehicle and the quote you were given, in whatever words the dealer used. You get an approximate fair-price range, the legitimate levers you can raise and an honest way to raise each, the questions worth asking, the charges to have itemised, the signals that mean you should walk away, and a short, polite opening line you can copy — one you can say truthfully. See the ethics section below; this is the part of the product with the firmest rules.
6. Dealer locator
Filter dealers by brand and city, or use “near me” to sort by distance from your current location. Each card shows the brand, city, address, a tap-to-call number and the distance when location is shared, with an optional map view. Dealer entries in the current build are sample listings for demonstration — confirm any dealer's details directly before travelling.
Plus: your garage & analytics
Signed-in buyers get a garage: every saved scan with its make, model, confidence chip and date, plus recent comparisons, searchable and one tap from the full detail. The analytics view adds the trend — vehicles scanned, comparisons run, the AI's average self-reported identification confidence, how many makes you have looked at, a scans-over-30-days line and a most-scanned-models bar chart. Average confidence is the model's own certainty about its own guess; it is not a measure of how often it was right.
Reading a result
Three things to look at before you trust anything on screen.
| What you see | What it means | What to do |
|---|---|---|
| Confidence (0–100) | How sure the model is about its own identification. A high number is not proof; a low number is a clear warning. | Below a comfortable level, correct the make and model yourself before running a cost or comparison on top of it. |
| Approximate range | A wide band, not a price. It moves with variant, city, taxes, offers and the day you walk in. | Use it to know roughly where you stand, then ask the dealer for the written on-road breakdown. |
| Assumptions | The inputs behind a cost estimate — running distance, fuel price view, service and depreciation assumptions. | Read them first. If an assumption does not match your life, the total does not apply to you — change the inputs. |
When the model cannot identify a vehicle or cannot give a sensible range, it says so rather than inventing one. An honest “we could not identify this” is a correct answer.
Negotiation ethics
The coaching feature exists to close an information gap, not to help anyone win by trickery. The rule the product is built around: a good deal is one both sides would accept knowing the truth.
What it will coach
- Genuine timing — month or quarter end, a model-year changeover
- Real competing offers you actually hold
- Unsold stock, or accessories and service instead of a discount
- Honest readiness to buy, said plainly
- Asking for every charge to be itemised in writing
What it refuses
- Inventing a rival quote you do not have
- Misrepresenting your budget or your intentions
- Pressure tactics and manufactured urgency
- Wasting a salesperson's time in bad faith
- Anything that only works if the other side is misled
The opening line the coach gives you is written to be something you can say truthfully; it never puts a claim in your mouth that might not be true. The screen says so, and asks you to edit it to match your real situation. The walk-away signals point the same way — refusal to itemise, terms changing at signing, pressure to sign immediately — because recognising bad faith is part of dealing fairly.
How it works
Point the camera at the vehicle and capture a frame, or upload a photo. The image is resized in the browser before it is sent.
A vision model returns its best guess — make, model, variant, body type, year range — with a confidence figure and typical specs. Correct it if it is wrong.
Enter how you will actually use the vehicle. The assumptions are shown first, then the cost heads, the total and the per-kilometre figure.
Compare your shortlist, prepare honest negotiation notes, find a dealer — then confirm every figure with them in writing before you commit.
Under the hood
Front end. ASP.NET Web Forms pages on a shared master page, Tailwind for layout, vanilla JavaScript and AJAX page methods for every data operation — no full page refresh for a scan, a calculation or a save. Skeleton loaders while the AI call runs, empty states everywhere, and the app installs as a PWA.
AI layer. A single helper class wraps every model call — identify, cost, compare, coach — each with a system prompt carrying the same responsible-AI framing, a strict JSON return shape, and a safe fallback so a failed call degrades to an honest “we could not do this right now” rather than a broken screen or an invented answer. Every call is logged with its duration and outcome.
Identity. Sign-in runs on the shared account kit scoped to product code ACB. The scan-to-shortlist path needs no account at all; anonymous activity is stored with an empty user field and counted against a free-scan limit held in the browser.
Data. All access goes through a parameterised data helper calling stored procedures. Every table carries soft-delete and audit columns; nothing is hard-deleted.
Plans & business model
- Scan and identify — no account needed
- A set number of scans to start with
- AR spec panel, cost of ownership, comparison
- Fair-deal coaching and the dealer locator
- Scans without the free-tier limit
- Saved garage — scans and comparisons
- Full personal analytics
- All scan, cost, comparison and coaching features
Billed monthly, no setup fee. Cancel any time.
The dealer-commission side
Alongside the consumer subscription, the product is designed to earn a commission when a buyer it referred goes on to transact with a listed dealer. This is worth being explicit about, because it is exactly the kind of arrangement that can quietly corrupt a buyer-side tool.
The line the product holds: commission must not influence advice. Today that is enforced by the architecture rather than by policy alone — and the distinction matters. The four advice methods (IdentifyVehicle, CalculateTCO, CompareVehicles, NegotiationCoach) are never passed a dealer identity, a listing, or any commercial term. They receive only the vehicle, the buyer's stated priorities and usage, and the quoted price. A dealer relationship therefore cannot reach an identification, a cost estimate, a comparison score or a negotiation note, because that data never enters the call. The dealer locator is a separate surface, and the coaching still tells a buyer to demand itemised charges and to walk away from a bad-faith deal — including at a listed dealer.
Not yet built, and required before any commission goes live: paid or commercial placements are not currently labelled as such in the dealer locator, because no listing is commercial yet — the seeded dealers are demo data. If and when a listing becomes paid, it must carry a visible commercial label, and ranking must not be sold. Until that control exists, do not present the locator as a neutral market survey.
Privacy & data
Anonymous to start. You can scan, cost, compare and get coaching without an account. Anonymous activity is counted locally against a free limit and is not tied to a personal profile.
Camera and images. The camera is only opened when you ask for it, and the browser will ask your permission first. When you capture or upload a photo, that frame is sent to the AI vision model to identify the vehicle. Only scans you explicitly save are stored against your account.
Location. Location is used only when you tap “near me” on the dealer locator, only to sort dealers by distance, and only with the browser's permission. Nothing is tracked in the background.
What we store. For saved scans: the identification, the confidence, the specs and the result. For comparisons: the shortlist, your priorities and the outcome. Enough to rebuild your garage — not a live camera feed.
Accounts & deletion. Registered buyers can delete their account from within the app and via a public deletion-request page that needs no login.
Payments. Premium payments are processed by the payment gateway; card details are never stored by the app.
Not a valuation. Identifications, prices, running costs, comparisons and negotiation notes are AI-assisted estimates for research. They are not a valuation, a quotation, financial advice or an authority. Confirm with the dealer or manufacturer.
Data model
Core tables (prefix ACB_). Every table carries soft-delete and audit columns.
| Table | Holds |
|---|---|
ACB_Users | Registered buyer accounts, with premium status and expiry. |
ACB_VehicleScans | One row per saved scan: the image reference, the identified make, model, variant guess, body type and year range, the confidence, and the full result. The user field is empty for anonymous scans. |
ACB_VehicleCache | Cached vehicle knowledge — typical specs and approximate price ranges per make and model — so repeat lookups need not re-ask the model. |
ACB_PriceData | Indicative price points by variant and city tier. Approximations for orientation, not quotations. |
ACB_ComparisonHistory | Saved comparisons: the shortlist, the buyer's stated priorities and the resulting matrix and recommendation. |
ACB_Dealers | Dealer listings: name, brand, city, address, phone and coordinates for the locator. Sample listings in the current build. |
ACB_Subscriptions | Premium subscriptions: plan, amount, dates and payment status. |
ACB_activity_log / ACB_error_log / ACB_ai_calls | Activity trail, error log, and a record of every AI call with its kind, duration and outcome. |
Identity uses the shared account kit scoped to product ACB. Every activity table stores the kit user id in its user field, or leaves it empty for anonymous use. The analytics view reads ACB_usp_GetAnalytics, which returns headline counts, scans per day for the last 30 days, and the most-scanned models — all scoped to the signed-in buyer.
Future roadmap
Decode a written quote
Photograph the dealer's printed quotation and have each line explained — what is statutory, what is a real cost, what is an optional add-on that has quietly been included.
Deeper EV modelling
Home versus public charging mix, tariff slabs and battery-life assumptions as explicit inputs to the cost model, so the petrol-versus-electric question can be answered against a buyer's own numbers rather than a generic average.
Used-vehicle inspection checklist
A guided checklist for a second-hand purchase — what to look at, what to ask for, which documents to verify — with photo capture against each item, and the same honest framing about what a photo can and cannot tell you.
3D model in your driveway
Place a scaled 3D model of a shortlisted vehicle in your own parking space through the camera, to answer the question a spec sheet cannot: does it actually fit?
Shortlist watch
Keep an eye on a saved shortlist and notify the buyer when something material changes — a variant update, a model-year changeover — with the same rule that any figure shown must still be confirmed with the dealer.
More languages
Spec panels, cost narratives and negotiation notes in more Indian languages, so a buyer can prepare in the language they will actually negotiate in.