Due to the increased interest in deep learning in recent years, there has been a significant upsurge in research in regard to the implementation of machine learning-based systems in production environments. The main points of interest has been in defect detection, categorisation and full automation.  We collaborated with Metaqlix AB, a company interested in enhancing quality assurance processes for their clients using image recognition technology, to create a product for a pharmaceutical production line. The technology uses deep learning, specifically the Faster R-CNN architecture, to find packaging defects in real-time while moving along a conveyor belt. To train the model, a dataset of images is created and manually annotated to highlight various packaging defects. The system is then evaluated for its accuracy and efficiency in detecting these defects. The results became increasingly more efficient with the latest accuracy rate reaching 88.33%, which in turn shows that the proposed system can effectively automate the defect detection process, which improves the overall quality assurance workflow in pharmaceutical production. Further discussion includes the details around some of the challenges that arose during the development process, including questions regarding data quality and how well the system performs under different conditions. By highlighting these aspects, the study provides valuable insights for future improvements and enables Metaqlix to better tailor its solutions to meet customer demands. In summary, this study shows a method in which machine learning and image recognition can change the quality assurance process in the pharmaceutical industry.