This paper examines the integration of Industry 4.0 technologies, including edge computing, spectroscopy, federated learning, and machine learning, to enhance cocoa bean quality inspection within the Malaysian bakery industry. The proposed solution addresses the urgent need for industry compliance with stringent international food safety standards. A federated vision-based framework employing edge-enabled learning is introduced to enable real-time, non-destructive product analysis, thereby supporting food safety and traceability. Using tools such as spectrophotometers, colour sorters and Orange Data Mining, cocoa beans are classified with high accuracy via machine learning confusion matrix evaluation while preserving data privacy. Validation of the method has shown a reduction in cocoa bean inspection errors to less than ten per cent. This study’s key technical contribution lies in its scalable, privacy-preserving quality inspection framework that integrates federated learning and edge computing to support decentralised learning. Practically, it offers a viable solution for Small and Medium Enterprises (SMEs) by reducing human error, improving inspection accuracy and enabling digital transformation in cocoa-based food processing. However, implementation challenges such as workforce skill gaps, limited access to labelled datasets and ethical data-sharing concerns must be addressed to support broader adoption across the agri-food industry.