In this research, a hybrid model based on artificial neural network and genetic algorithm was developed for optimization of the industrial production process of sodium co-silicate. The required data were collected from the production line during the years 2023 and 2024, and a total of 245 valid operational records were used. The input parameters included caustic soda consumption and inlet silica density, while the outputs included modulus, viscosity, and final product density. A multilayer perceptron neural network with a 2-10-10-3 architecture and hyperbolic tangent activation functions in the hidden layers and linear activation function in the output layer was trained using the Levenberg-Marquardt algorithm. The model performance was evaluated on the test data with coefficients of determination (R²) of 0.945 for modulus, 0.926 for viscosity, and 0.932 for output density, indicating the high accuracy of the model in predicting product quality indicators. The developed ANN model was then used as the objective function in the genetic algorithm with a multi-objective fitness function (including increasing modulus, increasing viscosity, and reducing caustic soda consumption). The optimization results led to the discovery of an optimal point in which, with a 2.7% increase in inlet density, caustic soda consumption decreased by 11.2%, while the product modulus increased by 8.7% and viscosity increased by 10.4. Implementation of the results on the actual production line for 7 working days and analysis of 21 samples showed a relative error of less than 1% between model predictions and practical results, confirming the high validity of the model.