Neural Network seems don't backpropagate really well

Hi! I'm trying to implement Deep learning to classify MNIST and it seems wrong and won't improving anymore. I try to implement all architecture by hand but seems it not learning really well. Any Help, Suggestion, or Hint might be helpful

Google Collab: [https://colab.research.google.com/drive/1OB-d6AMYdRFIPCPwxvv14gHtRdFJMphW?usp=sharing](https://colab.research.google.com/drive/1OB-d6AMYdRFIPCPwxvv14gHtRdFJMphW?usp=sharing)

Sorry for my selfishness, and thank you!

architecture deep learning deeplearning mnist network neural network