Inspiration
We started with a narrower question: can computer vision read embossed tyre text from phone photos? Early experiments showed detection and OCR were possible, but a fixed script wasn’t enough. Weathered rubber, rotated text, and noisy markings like RADIAL and TUBELESS break linear pipelines. True operational value needs agency that plans, retries, verifies against external sources, and knows when to ask a human.
What it does
It uses computer vision to locate embossed tyre-size text (e.g. 195/60R16), reads it through a multi-stage OCR pipeline with automatic retries when results are unclear, and validates the read against a product catalogue and each vehicle’s expected fitment in the fleet database. A unified confidence score decides the outcome: high-confidence matches are auto-approved and written to fleet records; ambiguous, mismatched, or unreadable cases are routed to a human review queue with a full reasoning trace.
Humans act only where judgment matters (approving, correcting, rejecting, or requesting a new photo) before anything is committed to the master fleet database. The result is faster compliance processing, fewer transcription errors, real-time fitment mismatch alerts, and an immutable audit trail for safety and legal accountability.
Log in or sign up for Devpost to join the conversation.