AI & ML enthusiast passionate about building scalable, production-ready intelligent systems

github

Vansh Patel  • Peerlist

LeafCure is an end-to-end AI-powered plant disease detection platform designed to help farmers, researchers, and agriculture enthusiasts identify plant diseases from leaf images. The application leverages Vision Transformer (ViT) models to classify diseases across 38+ categories, providing fast and reliable predictions through an intuitive web interface.The platform was built with a focus on production-ready deployment rather than just model training. I optimized model loading, inference latency, and request handling to ensure a smooth user experience while serving predictions efficiently. The backend exposes REST APIs using FastAPI, making it easy to integrate with other applications or mobile clients.To improve deployment performance, I profiled model execution, optimized inference workflows, and implemented efficient preprocessing pipelines for image normalization and transformation. The result is a scalable AI application that demonstrates how modern transformer-based architectures can be deployed for real-world agricultural use cases.

LeafCure – AI-powered Plant Disease Detection

I built an end-to-end Automatic Number Plate Recognition (ANPR) system to automate vehicle identification from images and video streams. The system detects license plates using YOLO, extracts the plate region, performs OCR to recognize the registration number, and automatically blurs the detected plate to preserve privacy when required.To improve recognition accuracy, I implemented multiple image preprocessing techniques such as grayscale conversion, adaptive thresholding, denoising, and contrast enhancement before passing the cropped plate to the OCR engine. These preprocessing steps significantly improved text extraction under varying lighting conditions and image quality.The project follows a modular inference pipeline where object detection, OCR, preprocessing, and post-processing are separated into independent components, making the system easy to extend and deploy. The architecture supports both image and video inputs and can be adapted for applications such as parking management, toll collection, traffic monitoring, and surveillance.