Released / PII detection for Nepali and English
रक्षाRakshya, personal data found in Nepali and English text.
Paste a message, a form or a news story, in Devanagari, in Nepali written in Latin letters, or in English. Rakshya marks the names, places, phone numbers and identity numbers in it, and it does so inside your browser.
Version 0.2, 14 November 2026
- 41 MB
- the whole model, small enough for a web page
- 96%
- of the personal data in Nepali news it had not seen, counted in characters
- 21
- kinds of personal data, from names to number plates
Free for research and other non-commercial use. It misses some personal data, so read the result before you rely on it.
Try it
The model and the rules run in this page. Your text is not sent anywhere.
Found
With it taken out
What it finds
- PERSON names of people
- LOCATION places and addresses
- ORGANIZATION offices, companies, parties
- DATE days, months and years, in both calendars
- PHONE mobile and landline numbers
- EMAIL e-mail addresses
- URL web addresses
- USERNAME handles and login names
- IP_ADDRESS IPv4 and IPv6 addresses
- CITIZENSHIP_NO citizenship certificate numbers
- NATIONAL_ID national identity numbers, Aadhaar, SSN
- TAX_ID PAN and VAT numbers, GSTIN
- PASSPORT_NO passport numbers
- LICENSE_NO driving licence numbers
- VOTER_ID voter card numbers
- ID_NUMBER employee, student, patient and other numbers of a person
- BANK_ACCOUNT bank account numbers and IBAN
- CARD_NUMBER payment card numbers
- UPI_ID payment addresses
- VEHICLE_PLATE number plates
- SECRET passwords, keys, tokens and one-time codes
It reads Devanagari, Nepali written in Latin letters and English, with digits of either kind. Hindi is experimental. What has a fixed form, such as a phone number or a number plate, is found by rules; names and the numbers that need their context are found by the model.
Benchmark
882 real texts in seven test sets that no system had seen: news, web pages, government pages and comments. The number is the F1 of exact spans over every kind of personal data in a set. Higher is better.
| System | Weights | News (ne) | Web (ne) | Government (ne) | Comments (ne, Latin) | News (en) | Web (en) | News (hi) | Mean |
|---|---|---|---|---|---|---|---|---|---|
| Rakshya 0.2 | 40M | 0.878 | 0.802 | 0.731 | 0.529 | 0.803 | 0.703 | 0.804 | 0.750 |
| Rakshya 0.2 server | 278M | 0.870 | 0.815 | 0.740 | 0.520 | 0.813 | 0.740 | 0.800 | 0.757 |
| Gopya 0.1 | 34M | 0.855 | 0.683 | 0.636 | 0.233 | 0.729 | 0.563 | 0.725 | 0.632 |
| EU PII anonymisation bards.ai | 278M | 0.647 | 0.601 | 0.409 | 0.253 | 0.787 | 0.530 | 0.727 | 0.565 |
| XLM-R base NER HRL Davlan | 277M | 0.660 | 0.617 | 0.267 | 0.302 | 0.772 | 0.498 | 0.712 | 0.547 |
| XLM-R large, CoNLL-2003 FacebookAI | 559M | 0.651 | 0.574 | 0.200 | 0.265 | 0.783 | 0.481 | 0.760 | 0.531 |
| EU PII anonymisation v3 bards.ai | 308M | 0.144 | 0.704 | 0.328 | 0.162 | 0.605 | 0.709 | 0.528 | 0.454 |
| Nym PII multilingual Wismut | 308M | 0.427 | 0.420 | 0.301 | 0.252 | 0.655 | 0.478 | 0.595 | 0.447 |
Of the 25 published systems we ran, none scores higher than Rakshya 0.2 on Nepali news, Nepali web pages, government pages, romanised comments, English news and Hindi news. Ahead of it is EU PII anonymisation v3 on English web pages (0.709 against 0.703).
Exact means that start, end and label are all right. The table has Rakshya, its first release and the 5 published systems with the highest mean. All 25 systems, the scores by kind and the method are in the full benchmark.
Use it
In Python, with no graphics card:
# pip install git+https://github.com/fiste-ai/rakshya
from rakshya import Rakshya
pii = Rakshya()
pii.redact("मेरो नाम सीता थापा हो, फोन ९८४१२३४५६७।")
# मेरो नाम [PERSON_1] हो, फोन [PHONE_1]।
From a shell, rakshya "some text" does the same. In a web page, load rakshya.js as the README shows; this page is the example. A larger model for servers is at fiste-ai/rakshya-server. Weights CC BY-NC 4.0, code MIT. It is a helper and not a guarantee: in Nepali news it misses about 4 in 100 characters of personal data, and more on text that is unlike news.