Artificial vision systems use complex algorithms to identify specific characteristics in the objects of study, but by integrating artificial intelligence, the accuracy and simplicity of the processes increases. The project focuses on a scalable, versatile and lightweight prototype, resulting in an automatic candling system. Three main characteristics of chicken eggs, used in pastry and pastillage, were identified, classifying them as normal, cracked and with excess pores. In addition, a prototype was developed with a wooden structure that houses a Smartphone in charge of capturing images, LED panels for illumination and a special perforated cuvette. The captured images are transmitted to a laptop via WiFi and processed with Python algorithms using libraries such as Qt, OpenCV, Keras and TensorFlow. Two scripts manage the training of the neural network model and the system interface, which monitors, controls and stores data in databases managed by MariaDB and phpMyAdmin. With 900 images (720 for training and 180 for validation) and 2700 additional images analyzed, an accuracy of 93.93% has been achieved with a response time of 0.021954 s per egg. The system is suitable for candling of any oviparous egg.