Neural networks and data science become more and more popular. Though regardless of the late appearance of new advanced libraries permitting neural network creation and training seemingly “in one code-line”, there remains a starting-point problem, when new-coming pupils cannot understand the subject before they create the working code, but they see an interface of black box laboratory works. The author having dealt with universal and advanced applied mathematics courses practiced a clear and concise neural networks introduction, reflecting state of the art before the 2012 year “Big Bang” when current top-used convolution, LSTM, and other Transformer-like networks become super-effective and super-popular. Instead, the presented course consists of approximately 5 of the very classical labs adopted to mainly standard Excel (without VBA) and in 1-2 cases to C#, and, more important we solve a problem of building a neural network or a model from the white sheet (including corresponding experiments) in a standard 4 academic hour time.