25 Years of Eggs
Summary
This post is a detailed case study of using a hybrid AI/ML pipeline to extract and quantify egg-related purchases from 11,345 receipts collected over 25 years. It covers OCR challenges, tool comparisons (Tesseract, PaddleOCR-VL, SAM3 segmentation, and large language models like Codex and Claude), pipeline architecture, ground-truth labeling, and the resulting data quality and economics. The narrative offers practical lessons on building end-to-end data extraction workflows with attention to reliability and scalability.