Coffee is a popular beverage with a lengthy production process from planting to serving. Unfortunately, defects can persist during the post-roast stage, and the machines specializing in sorting can be costly or unavailable. This study utilized roasted coffee beans’ primary (RCB1) sourced from Davao City, Philippines and secondary (RCB2 and RCB3) online datasets with varied preprocessing techniques including image augmentation. The object detection algorithms YOLOv5, YOLOv8, R-CNN, and Detectron2 were evaluated through their precision in classifying post-roast defects and roast levels. The algorithms were trained and evaluated on large, labeled image datasets for accurate post roast defects and roast levels classification. The results showed that the precision for the RCB1 dataset yielded 79.8% for YOLOv5 and 92.5% for YOLOv8, whereas the precision for the RCB3 dataset was 49.4% and 97.6% respectively. The precision for the RCB2 dataset was 99.3% for YOLOv5 and YOLOv8, 98.4% for R-CNN, and 71.0% for Detectron2. The YOLOv8 was deployed for the lightbox using camera and has shown potential for efficient roasted coffee sorting. Varied datasets and cross platform applications should be considered for future work.