Plates read at the edge,
verified here.

Every vehicle is detected, read and classified on the device itself. What the pipeline is unsure about lands in a review queue — and each correction becomes training data for the next model.

• On-device detection, OCR and classification
• Store-and-forward — reads survive network and power loss
• Human review loop that feeds model fine-tuning