Spending categories in a budgeting app

Design the part of a personal budgeting app that sorts each card transaction into a category such as groceries, dining, rent or transport. The app has 300,000 users and imports about 4 million transactions a month from a bank data provider in daily batches. A trained classifier predicts a category from the merchant text, amount and merchant category code. Users can correct a category, and a correction should apply to that merchant for that user from then on. Describe when transactions are categorized, how user rules and the model work together, and how corrections are used to improve the model.