Applying Drivetrain approach on a project

Hi everyone, I just completed lesson 2 of part 1. I am very impressed with the drivetrain approach and wanted to use it to create an image classifier for persons ringing the doorbell at my home. Can you all please let me know how I can improve my drivetrain approach in solving it? Below is the one that I created

  1. classifying people who ring the bell of the home
  2. category of people according to different examples given, including relatives, delivery agents, house helpers, etc.
  3. collecting images of people for a week or two who arrives at the front gate and collaborating images of relatives from photo albums
  4. using the deep neural net to build the model.