CarScan AI

Snap any car.
See it in stock.

Photograph a car and AI identifies the make, model and year - then checks it against a dealer inventory of 177 vehicles.

Or try a sample
Under the hood

How it works

A four-stage pipeline running on a single serverless function. No app to install, no server to keep alive.

01 · Capture
Browser camera The photo is resized to 1280px in the browser before upload, keeping it quick on mobile data.
02 · Identify
Vision model Claude reads the badge, grille and lamp signature and names the vehicle - without being shown the catalog first.
03 · Match
Catalog lookup Only then is the result matched against inventory. No match returns nothing - never the nearest lookalike.
04 · Price
Live stock Matching units come back with trim, mileage, price and availability, newest and cheapest first.
VisionClaude Sonnet 4.5 · temperature 0 · forced JSON contract · two-stage prompt separating identification from catalog matching
RuntimeOne Vercel serverless function · static front end · no framework, no build step
LimitsRedis-backed per-visitor and global daily caps, so a viral moment can't run up an API bill
The inventory is fictional - generated for this demo, not a real dealership. The identification is real.
Sample photos from Wikimedia Commons (CC BY-SA / CC0). Uploaded photos are sent to the vision model for identification and are not stored.