I’m developing a smart dustbin inspired by the Ameru AI bin and I need a complete vision-based sorting solution that runs on an NVIDIA Jetson Orin Nano. The system must recognise and divert organic, recyclable and electronic waste (plus anything that doesn’t fit those categories) into separate compartments in real time. Your job starts with designing and training the computer-vision model, then packaging the inference pipeline with TensorRT/DeepStream—or whichever framework you prefer—for low-latency execution on the Orin Nano. After that, integrate the model with our bin’s actuators so the correct flap or chute opens automatically. Beyond sorting, the bin will feature an active odour-control module and a Bluetooth/Wi-Fi link to a companion mobile app. When onboard weight or fill-level sensors detect the bin is full, the firmware should push a mobile notification through the app’s API. All control logic must be documented so my hardware team can wire the sensors and deodoriser around your software stack. Deliverables • Optimised CV model and inference code for Jetson Orin Nano • GPIO/servo control scripts linked to classification output • Odour-control and fill-level sensor routines with API hooks • Mobile-app notification integration (SDK or REST) • Step-by-step setup guide, including flashing and dependency install If you have prior waste-sorting or edge-AI experience, especially with YOLO, EfficientNet, or similar networks on Jetson boards, I’d love to see it. Please outline your proposed approach, expected accuracy, and a realistic timeline so we can move quickly to hardware prototyping. For better understanding please refer to this https://youtube.com/shorts/WiNRz2VhiV4?si=6sXxnQtbLVAco_gA.