Side ProjectAI-Built

CarBecho: Used-Car Inspection Co-Pilot

CarBecho is an interactive inspection flow for used-car field inspectors: job list → verify & pair → a 200-point checklist (8 sections, 40 rows, Yes/No only) → report. An AI chat co-pilot (Gemini) is available from every row, accepting text, voice, or photo answers and fuzzy-matching replies straight onto the checklist row with a one-tap Mark button. A companion voice-only mode lets an inspector just ask a question and get a spoken answer. Findings feed a shared log with search and severity/section/photo filters. Built with Next.js and Supabase, directed end-to-end via Claude.

From a PM's Desk

Who it's for: Field inspectors running used-car inspections against a 200-point checklist, usually one-handed, often mid-conversation with the seller — UI friction here isn't an annoyance, it's the reason the checklist doesn't get finished.
Approach: Bet that a multimodal co-pilot (voice/text/photo, fuzzy-matched to checklist rows) beats a faster form. The cheaper alternative — just speed up the form — doesn't fix the real constraint: inspectors can't type mid-inspection.
Metric I'd chase: North star: time to complete a full inspection. Guardrail: co-pilot match accuracy — if inspectors correct the AI more than they'd have just tapped manually, the bet is wrong.
Status: Live prototype.

View →