In this workshop, Gretel and Google showcase an end-to-end machine learning approach to automatically anonymize sensitive customer data in financial documents.
We use synthetic customer data to teach a small language model how to anonymize sensitive fields like names, account numbers, transactions details, etc. This avoids manual data collection and labeling, and can be easily adapted and customized to support customers’ specific data patterns. The end result is an automated, customizable system for anonymizing financial data that improves over time.
Key topics covered include:
- Generating synthetic training data
- Fine-tuning language models
- Deploying through a Streamlit application
- Enabling human-in-the-loop validation
