Develop an End-to-End solution for the Handwritten Digit Recognition System ( MNIST Data : https://keras.io/api/datasets/mnist/ ) . The system should have the following specifications

1. It should have an ML model as the core program API - you can find models ( code ) built on TensorFlow or PyTorch easily and use them.

2. The solution should have a frontend to display results - a simple upload image button, where the user uploads an image from the computer and hits predict, and the solution shows the predicted number with the uploaded image.

3. The solution should have a backend ( database connection ) - for simplicity if I upload an image with the same file name as has been uploaded before, the system should not be predicting, it should simply give the output from a stored answer derived from predicting before, while printing on the screen : “Image predicted before“

4. The final output can be a recorded video explaining the system build and displaying results as required above