Developed a custom neural network from scratch in Python, without relying on external machine-learning libraries. Implemented the full feed-forward and backpropagation algorithms manually, including weight updates and error minimization. Designed a network architecture with 1 input neuron, 3 hidden neurons, and 1 output neuron, using the Sigmoid activation function. Created a reusable and scalable weight structure using a single indexed weight array (w[0], w[1], …`). Implemented mathematical functions such as the Sigmoid function and its derivative for training and gradient calculations. Built a supervised training loop capable of learning a target function (e.g., approximating ln(10) from input=10). Performed data loading and preprocessing using pandas, including initial work with the Sonar dataset. Gained strong hands-on experience with neural-network fundamentals, weight optimization, activation behavior, and low-level ML architecture.