Feryadi Yulius

A computer vision project for detecting and monitoring pineapple plants and fruits from images and videos. The system explores object detection, confidence visualization, and a pipeline for future tracking and counting in precision agriculture.

Bachelor thesis evaluating Faster R-CNN and YOLO for object detection on coastal water imagery. The work compares accuracy and inference speed through preprocessing, augmentation, fine-tuning, and evaluation using mAP, precision, recall, and F1-score.

A YOLOv8 computer vision model for real-time pothole detection and spatial localization to support road maintenance planning. The project achieved Precision 0.831 and mAP@50 0.774.

Processed 236 images across 26 phytoplankton classes, expanded the dataset to over 10,000 images using nine augmentation techniques, and evaluated 13 CNN architecture variants. The work was deployed as a website and Android app, PlanktoScan, with BRIN copyright registration.

A quantitative study of vehicle queues at gas stations near Institut Teknologi Sumatera using M/M/1 and M/M/s queuing theory. The analysis identified peak congestion at ρ = 1.14 and recommended adding one server during rush hours.

A personalized outfit recommendation system using Retrieval-Augmented Generation with NVIDIA Llama-3.1 Nemotron Nano 8B. It uses FAISS similarity search and prompt engineering to create context-aware stylist responses.

An LSTM autoencoder that generates text embeddings from company profiles for similarity search. The work evaluates semantic similarity through embedding analysis and PCA/t-SNE visualization.

An infographic titled “From Fields to Distribution” that uses harvest area, average grain prices, and total production data to highlight geographical inequalities in Indonesian food security. Submitted to the 2025 Gammafest IPB Statistical Visualization Competition.

A business intelligence dashboard for Maven Toys that tracks total orders, revenue, profit, product-category trends, and revenue growth for interactive management analysis.

An AI-powered platform that identifies 31 varieties of Indonesian spices through image analysis. The Streamlit app reaches up to 83% prediction accuracy and supports cultural and culinary education.

An interactive website that detects batik types from uploaded images. It applies image processing for pattern recognition and supports digital cultural preservation and public education.