Back to projects
FeaturedCompleted

NumPy Neural Network Engine (From Scratch)

A custom-built, lightweight neural network library implemented from first principles in Python/NumPy, without external deep learning frameworks.

PythonNumPy

Designed and implemented a multi-layer perceptron (MLP) from scratch to demonstrate a deep understanding of machine learning fundamentals. The project required manual implementation of matrix-based forward propagation, backpropagation using the chain rule, and stochastic gradient descent for weight optimization. By bypassing high-level abstractions, this project provides insight into the exact mechanics of neural architecture, activation functions (ReLU/Sigmoid/Softmax), and the mathematical rigor required for efficient model training.