Engineered end-to-end real-time computer vision pipelines for a parking and toll automation provider serving 20+ airports and 60+ malls across India. Built distributed microservices using Deepstream 5.1 with dual deep learning models (detection + OCR), achieving 25+ fps on cost-efficient Jetson Nano edge devices deployed across 80+ live sites.
Architected dual-model ML pipeline optimized for accuracy and speed, then created synthetic data generation tools that produced 50M+ labeled training images, enabling robust performance across diverse lighting conditions and vehicle types. Designed MQTT-based IoT coordination (Paho/Mosquitto) for multi-stream video analysis, synchronizing gate and barricade triggers in real-time.
Developed Django REST APIs exposing OCR results to third-party parking automation systems, enabling seamless integration with existing infrastructure. The modular microservices architecture scaled across dozens of simultaneous video streams while maintaining sub-second latency for mission-critical triggers.
Technologies: Python, Deepstream 5.1, TensorFlow, Keras, Jetson Nano, MQTT, Django, REST APIs, Computer Vision, IoT.