Skip to content

Find and mask personal identifiers

pii finds Indian mobile numbers, email addresses, URLs, Aadhaar numbers, PAN codes, and PIN codes. It reads Odia digits and ASCII digits, and it runs offline. It does not find names, street addresses, dates, or amounts, so masked text can still identify a person.

pii needs a release newer than 0.1.14. On PyPI 0.1.14, import openodia.pii fails and openodia pii exits with code 2. From a source checkout, uv run python can import it.

Every example on this page is run by the test suite.

Find matches

detect returns one PIIEntity per match, in order. start and end are character offsets into the text you passed.

>>> from openodia import pii
>>> pii.detect("ରାମ ସୀତା ଘରକୁ ଯାଏ ।")
[]
>>> text = "ମୋ ଫୋନ ୯୮୭୬୫୪୩୨୧୦ ଓ a@b.com"
>>> pii.detect(text)
[PIIEntity(type='PHONE', text='୯୮୭୬୫୪୩୨୧୦', start=7, end=17, confidence=0.9), PIIEntity(type='EMAIL', text='a@b.com', start=20, end=27, confidence=0.95)]

A 10-digit number that starts with 6, 7, 8, or 9 is a PHONE, including an order number that happens to have that shape. A PIN code is reported only after PIN, Pincode, or ପିନ.

>>> pii.detect("PIN 751001")
[PIIEntity(type='PINCODE', text='751001', start=4, end=10, confidence=0.8)]
>>> pii.detect("751001")
[]

Mask matches

mask replaces each match with <TYPE>. With those tags, masking the result again changes nothing, unless the input chains more than 10 values that were glued together.

>>> pii.mask(text)
'ମୋ ଫୋନ <PHONE> ଓ <EMAIL>'
>>> pii.mask(pii.mask(text))
'ମୋ ଫୋନ <PHONE> ଓ <EMAIL>'
>>> pii.mask(text, types=["EMAIL"])
'ମୋ ଫୋନ ୯୮୭୬୫୪୩୨୧୦ ଓ <EMAIL>'

types is a list of PHONE, EMAIL, URL, AADHAAR, PAN, and PINCODE. A bare string raises TypeError.

>>> pii.detect(text, types="EMAIL")
Traceback (most recent call last):
  ...
TypeError: types must be a list of PII type names, not a single string; wrap it in a list

Drop a weak Aadhaar match

An Aadhaar-shaped number that passes the Verhoeff checksum has confidence 0.95. One that fails has confidence 0.5. min_confidence drops weaker matches before overlapping matches are merged.

>>> pii.detect("234567890124")[0].confidence
0.95
>>> pii.detect("234567890125")[0].confidence
0.5
>>> pii.mask("234567890125", min_confidence=0.9)
'234567890125'

Limits

Overlapping matches become one span. The type is the longest match. When the lengths are equal, the earlier of EMAIL, URL, AADHAAR, PAN, PHONE, and PINCODE wins, so mask does not leave part of a detected value visible.

mask repeats until the text stops changing, and it stops after 10 passes. After more than 10 glued values, the text it returns can still contain a raw value.

An email local part longer than 64 characters is not found. When that local part contains punctuation, only its last 64 characters may be masked. A local part that starts with Odia letters is masked from the first ASCII character, so the Odia letters stay visible.

detect and mask raise ValueError for an unknown type, or for a min_confidence outside 0 to 1.

The pattern for each type is in PII detection and masking. The command is in Command line. openodia pii --json includes the raw matched text. Use --mask when the output should hold only the masked text.

Next