AI Banking Revenue Categorizer

I have a continuous feed of raw banking-transaction exports (CSV and JSON) and I need an AI model that can examine each record, run transactional analysis, and automatically label every incoming credit with the correct revenue category.

The workflow I picture is straightforward: your code ingests the files, cleans and normalises the fields, then applies a machine-learning or deep-learning model that decides whether the money is, for instance, product sales, service income, refunds, interest, or any other class we agree on. I am only interested in revenue-side categorisation; expenses can be ignored for now. Accuracy matters more than speed, but the system must still process a typical daily batch (≈10 000 lines) in minutes, not hours.

You will receive several months of historically tagged transactions to train and validate the model. I am comfortable with Python and would like well-commented scripts that rely on common libraries such as pandas, scikit-learn, TensorFlow or PyTorch, plus a concise README that lets me reproduce your results on my own machine.

Deliverables:
• Clean, runnable code (model training + inference)
• Trained model weights or checkpoint
• README with setup, execution steps, and metrics achieved on the validation set

Acceptance criteria: F1-score ≥ 0.90 on the supplied hold-out data and clear, reproducible instructions. Once this is met, I will integrate the categoriser into our wider reporting pipeline.