Airbnb uses Presidio and GLiNER for text de-identification

airbnb.tech11 min read

Airbnb details its architecture for de-identifying text at scale to protect personal data before training machine learning models. The system relies on Microsoft's open-source Presidio library combined with custom recognizers, zero-shot NER models like GLiNER, and consistent faker-based replacements. Airbnb applies this setup as a shared workflow step to ensure data privacy without sacrificing model utility. Open article →