Instead of gradient descent, each layer solves for optimal weights directly via ridge regression in a single forward pass. Works as a warm start before Adam — same final accuracy, fraction of the total training time.Repo: https://github.com/infiplexity-pixel/to_the_point/