GDPR, PCI, PII - AI Driven

Any data collected by Germain UX can be masked, anonymized, or excluded (i.e., not collected) by Germain UX.

Germain UX automatically detects information considered sensitive, such as PII (Personally Identifiable Information). And that works for any KPI.

This section lists all data that is not yet hidden and should likely be set as private (either anonymized, masked, or excluded).

GermainUX automatically detects PII in the data that it collects in real-time and relies on BERT-based AI models for that purpose.

Our findings on other models:

ModelName	Based on	License	Params	Entity Names for person/address	Impression	Notes	Evaluated formats
1-13-am/deberta-pii-finetuned		mit	184M	NAME_STUDENT, STREET_ADDRESS	good	aggregation_strategy='first' Whitespace not preserved
ctrlbuzz/bert-addresses	bert-base-cased	?		PER, addr	good	Only US addresses	text, JSON
h2oai/deberta_finetuned_pii		mit		FULLNAME, FIRSTNAME, MIDDLENAME, LASTNAME, STREETADDRESS	good	aggregation_strategy='first'	text, JSON
jammmmmm/pii	lakshyakh93/deberta_finetuned_pii	mit	?	FULLNAME, FIRSTNAME, LASTNAME, STREETADDRESS, USERNAME	good	aggregation_strategy='first'. Attempts to detect username/password	text
lakshyakh93/deberta_finetuned_pii		mit	?	FULLNAME, FIRSTNAME, LASTNAME, STREETADDRESS	good	aggregation_strategy='first'
kulkarni-harsh/address-extraction-ner	distilbert/distilbert-base-cased	mit	65M	LABEL_1+LABEL_2	limited	Only addresses, not persons
yonigo/distilbert-base-multilingual-cased-pii		apache-2.0	135M	BUILDING	limited	Only addresses, not persons
zmilczarek/pii-detection-roberta-v2			124M	NAME_STUDENT, ID_NUM	limited	aggregation_strategy='first' Only the street number and zipcode of an address
DioulaD/birdi-finetuned-ner-address-v2	camembert/camembert-base				poor
hewonty/longformer-ner-finetuned-pii		apache-2.0	148M	NAME_STUDENT	poor

Service: Management

Feature Availability: 2024.3

Feature