Inspired from a repo . i have implemented an some ground level maths and logic . In this neural network no tensorflow , pytorch is used it is just prepared by using numpy and some logical math.

A dynamic neural network built from scratch using only NumPy — Personal project for learning purpose. No frameworks, no abstractions. Implements forward propagation, backpropagation, and batch gradient descent manually. Supports any user-defined architecture, binary and multiclass classification, and pickle-based weight persistence.