How we helped a Japan-based used-vehicle distributor turn a senior inspector's gut judgment into a standardized AI workflow — running on every Android and iOS phone in the field, online or off.
01 — EXECUTIVE SUMMARY
THE CLIENT
A used-vehicle distributor operating across Japan's auction and trade-in network.
THE CHALLENGE
Repair-history judgment was tied to a handful of senior eyes — and didn't scale to small branches.
THE SOLUTION
A Flutter app that guides 40-point capture, runs Gemini Vision on each frame.
KEY METRIC
~3× faster inspections — and a junior inspector now produces a report a senior would sign off on.
02 — THE PROBLEM
The client's quality of judgment lived in the heads of a small group of senior inspectors. Every branch without one of them, every online trade-in form, every batch of vehicles cleared by a junior — all carried hidden risk that only surfaced at auction.
Smaller dealerships couldn't staff every site with a senior. Junior inspectors made calls they weren't trained to defend.
When the customer submits photos, there's no senior in the room. Repair history was a guess until the car physically arrived.
Most cars don't need a 30-year career to clear. But seniors were spending their day on them anyway, instead of the cases that actually needed expertise.
COST OF DOING NOTHING — A
Mispriced inventory at auction. Margin loss compounds across thousands of vehicles per quarter.
COST — B
Trust erosion with downstream buyers when a "clean" car turns out to have history.
COST — C
A senior workforce nearing retirement, with no scalable way to transfer the judgment to the next generation.
03 — OUR APPROACH
Before a single line of code, our team spent days on inspection floors with senior staff — watching the order they walked around the car, the photos they actually took, the conditions where their judgment got blurry. That walk-around became the spine of the product.
Field shadowing
Two researchers, three inspection sites, six full inspections observed end-to-end. Every photo, every "skip," every disagreement logged.
Codifying the walk-around
We turned the senior's instinctive route into a fixed 40-point sequence — front, side, rear — with stable IDs that survive across capture, AI, and 3D rendering.
Designing the failure mode
The hardest design call: what does the AI say when it isn't sure? We chose indeterminate as a first-class status — never "OK" by default, never "NG" by guess.
Prompt + threshold engineering
We engineered structured prompts (point name, type), pinned temperature at 0.2, and set a confidence floor at 70 — anything below is automatically downgraded.
Validate against the seniors
Same vehicles, judged independently by a senior and by the app. We tuned the model and the UI until the disagreements were the right disagreements.
04 — THE SOLUTION
A Flutter mobile app that turns the 40-point capture, the AI judgment, and the report review into a single linear flow. Inspectors can capture in zero-bar basements; analysis runs the moment connectivity returns.
F.01 STRUCTURED 40-POINT CAPTURE
Front · side · rear, in the same order an experienced inspector would walk it. Skip when a clean shot isn't possible, Jump back to fix one — and a free-form note per point that lands in the final report.
F.02 AI JUDGMENT WITH CONFIDENCE & REASONING
Each photo lands at Gemini Vision with a structured prompt. We get back a status (pass / fail / indeterminate), a 0–100 confidence, and bilingual reasoning. Anything below 70 is auto-downgraded — the system never bluffs.
F.03 INTERACTIVE 3D CONDITION REPORT
All 40 results plotted as colored markers on a 3D vehicle model. Four preset views, free rotate / pinch / zoom. Tap any dot to see the photo, the confidence, the AI reasoning, and the inspector's note — all in one place.
F.04 BILINGUAL END-TO-END
JA / EN with one tap. Inspection-point names are stored bilingually in the model; AI reasoning comes back in both languages. The same report is reviewable by a Japanese-speaking and English-speaking buyer, no re-analysis.
05 — TECH STACK
Before a single line of code, our team spent days on inspection floors with senior staff — watching the order they walked around the car, the photos they actually took, the conditions where their judgment got blurry. That walk-around became the spine of the product.
MOBILE
Flutter
One codebase, iOS + Android, native camera access.
VISION AI
Google Gemini
Structured prompts, JSON output, temperature 0.2.
3D RENDERING
Three.js
Embedded in WebView — markers, presets, free rotate.
LOCAL STORE
SQLite
Offline-first capture; AI runs on reconnect.
"A junior with this app produces a report I'm willing to sign off on. That's the part nobody believed was possible."
SENIOR INSPECTOR, CLIENT OPERATIONS
Tokyo · 30 years on the floor
06 — LET'S BUILD YOURS
We work with operators in mobility, fintech, retail and logistics to turn senior judgment into software the whole company can use. Tell us what you're trying to scale — we'll come back with where to start.
Book a 30-min consult
LOGIQUE 
