Pre-harvest yield estimation for bird's eye chili often relies on visual judgment, which can be inconsistent when fruits are small, numerous, overlapping, or partly occluded. This study developed an end-to-end web-based workflow integrating YOLOv8s visible-fruit detection, Slicing Aided Hyper Inference (SAHI), category-weighted digital sampling, and harvest-mass conversion using mean fruit weight and plant population. The main contribution is not a new detector architecture, but a documented workflow that connects visible fruit counts, representative crop-condition proportions, and quantitative mass estimation. The model was trained on 855 field images of Ori 212 chili plants. After tiled preprocessing and retention of original images, 1,710 resources were used, consisting of 1,196 training, 342 validation, and 172 test images. Internal evaluation produced 0.89460 precision, 0.84559 recall, a 0.86 F1-score, 0.91064 mAP@0.5, and 0.55609 mAP@0.5:0.95. External counting evaluation on 60 plants from a different field was unbalanced across categories (24 low-, 24 medium-, and 12 high-density plants) and recorded 1,903 manual counts and 1,566 detections, with 17.43% mean absolute counting error. Limited mass validation used six plants, namely three H-14 and three H-1 samples. The H-1 samples produced 113.68 g estimated aggregate weight versus 127 g actual weight, corresponding to 10.49% error. An 8,000-plant scenario estimated 291.65 kg, but it was an illustrative formula-execution scenario based on user-entered population and category proportions rather than independently measured full-harvest ground truth. The system supports preliminary visible-fruit-based pre-harvest planning, while reliability depends on image visibility, sampling representativeness, fruit-weight measurement, and complete-harvest validation.