Pengembangan Sistem Pemilahan Sampah Dengan Manajemen Terintegrasi Menggunakan Mekanisme Penalaran Kontekstual Berbasis Visual Understanding Dan Generative AI

Modern waste management requires innovative digital solutions to increase community participation and material quality. This research develops a hybrid AI ecosystem for a mobile-based Smart Bank Sampah platform, integrating YOLOv8 Nano for real-time edge detection and Llama-3.1-8B-Instant for cloud-based cognitive reasoning. Unlike traditional systems, this framework converts visual data into personalized educational instructions, ensuring waste is properly prepared before transaction. Results show that the YOLOv8 model achieves a mAP50 of 95.4% with a rapid inference time of 15–25ms on mobile devices. The system successfully generates structured contextual guidance, validated by a 100% success rate in functional Black Box testing. Furthermore, usability testing via the System Usability Scale (SUS) yielded a score of 92.17, placing the application in the Best Imaginable (Grade A) category. This study demonstrates that combining Computer Vision and Large Language Models (LLMs) significantly improves administrative transparency and environmental literacy, providing a robust infrastructure for a sustainable circular economy