I need a small-footprint, trainable email classifier that can look at the subject line, sender information, and full message content, decide whether each message is spam or ham, and then act accordingly. Whenever the model decides a message is spam, it must flag the item for review and keep it out of the standard inbox so the user never sees it unless they actively inspect the flagged list. Please build and train the model in Python—feel free to use scikit-learn, TensorFlow, or another well-documented NLP framework—as long as the final solution runs reliably on a typical server environment. I will provide a starter dataset; leave the pipeline open so I can add more labelled examples over time and watch the accuracy improve. Deliverables • Clean, commented source code for data preparation, model training, and prediction • A brief README explaining how to retrain the model on new mailboxes • A test script that demonstrates the “flag for review” workflow (spam stays hidden, ham goes straight to the inbox) Acceptance criteria • Minimum 95 % precision on the validation split for the spam class • The script correctly routes and flags messages exactly as described above without deleting anything automatically