Built an OCR pipeline combining OpenCV preprocessing, Tesseract/Deep learning OCR, and post-processing with regex/entity validation. Deployed as a containerized microservice with API, monitored accuracy and error types for continuous improvement (-35% errors).