NerReplace is a Teradata Vantage BYOM Java Table Operator (UDF) that detects entities in text and replaces selected entities in the output. It is designed for in-database masking and redaction of sensitive information, such as Personally Identifiable Information (PII), using an ONNX-based Named Entity Recognition (NER) model and associated model artifacts.
NerReplace uses the following model components:
- ONNX model (stored as a BLOB in a ModelTable)
- tokenizer.json (stored as a BLOB in a TokenizerTable)
- config.json (stored as a BLOB in a ModelConfigTable)
When to use NerReplace
Use NerReplace when you want to detect and replace sensitive information directly within Teradata Vantage, avoiding unnecessary movement of raw text data outside the database.
This approach supports privacy-first data preparation workflows by keeping entity detection and redaction close to the data.