Ashwanthkk/Deep-learning-framework ? reverse-engineered prompt

Build me a small deep learning framework from scratch in Python using only NumPy, focused on convolutional neural networks.

I want it to show how convolution works internally, including forward propagation and backpropagation for Conv2D layers, with ReLU and a simple mean squared error loss, plus a basic gradient descent optimizer. Keep it educational and easy to read, not a full production library. Please make the code clean enough that someone can follow the math and see how gradients move through the convolution filters.

If it helps, include a tiny example or demo showing how to create a model, run a forward pass, compute loss, and update weights. Keep the scope honest since this is only for convolution layers right now, and don’t depend on TensorFlow or PyTorch. If you need current NumPy docs or best practices, look them up online.