Jinsith Ashokan

At AutoPlant System India Pvt Ltd, I prepared and labelled UAE number-plate images and fine-tuned Qwen2-VL for Arabic and English plate reading. I worked through difficult Arabic emirate text until the results were reliable.

I trained and evaluated a YOLOv8s detection model, achieving 99.5% mAP@50, and traced a per-class drop to zero detections to unnormalised calibration data.

I built a video object-counting system using YOLOv8s/YOLOv11l, SORT tracking and dual-ROI logic on Jetson Orin Nano. I also built a truck-plate ANPR pipeline that combines YOLO detection, SwinIR upscaling and vision-language OCR.

Previously, at Datamoulds Pvt Ltd, I built an ANPR system using YOLOv5 and PaddleOCR, along with a document-scanner backend. I’m seeking part-time remote work in image/video annotation, annotation QA and computer vision model evaluation alongside my full-time role.