Privacy Protection
Runs on incoming messages
Finds personal details in what users send your assistant, and removes them before the message is answered or saved.
Enabling this guard helps your organisation stay compliant with India's DPDP Act and GDPR.
What happens
↻ Fixes & passes on — The message continues as normal, with personal details hidden. The user is not interrupted.
Example
| User | Mera naam Rahul Sharma hai, number 98765 43210 hai aur PAN ABCDE1234F hai. |
| Assistant sees | Mera naam XXXXX XXXXX hai, number XXXXX XXXXX hai aur PAN XXXXXXXXXX hai. |
The real numbers are never saved or shown anywhere.
Without this guardrail
Personal details get saved and can be shown again in the chat, exactly as the user typed them. Risk: personal data can leak, and users stop trusting your service.
How well it works
| Accuracy | Response time |
|---|---|
| 86% | 0.02 sec |
Tested on 419 real-style messages in English, Hindi and Hinglish.
Technical details
| Method | Accuracy | Precision | Recall | F1 | Sample size |
|---|---|---|---|---|---|
| Rule + ML | 84% | 92% | 80% | 86% | 419 |
Dataset: Evaluation dataset (419 samples) ↗
Named-entity recognition tuned for Indian PII formats (Aadhaar-like numbers, phone numbers, addresses), evaluated on 419 real-style samples in English, Hindi and Hinglish. Precision (92%) means false alarms are rare; recall (80%) means about 1 in 5 real personal details slipped through unmasked. F1 (86%) balances the two.