Skip to content
TEKVO / PRIVACY & DATA GOVERNANCEPRODUCT IN DEVELOPMENT · PILOT DISCUSSIONS
TekvoPIIEPII Discovery Engine
Multilingual Personal Data Discovery

Find personal data across languages and scripts.

Review personal data in regional documents, bilingual forms and mixed-script records within your own environment.

For Government, privacy and document operations teams

Discuss this workflow ↗
THE WORK TO BE DONE

Start with the records your team needs to understand.

Records may combine English headings, regional-language fields and identifiers written in local numerals. Discovery needs to account for the script, the data category and the quality of the source.

STEP 01

Choose the language mix

Identify the languages, scripts and document types your team handles, including bilingual records.

STEP 02

Read and detect locally

Use configured OCR routes and local detection models, with numeral normalization before identifier checks.

STEP 03

Review in source context

Check candidate personal data against the original page and validate coverage using representative records.

LANGUAGE COVERAGE

Match language support to your records.

Coverage is assessed by task. The current repository includes separate evaluations for reading documents and detecting numbers, with model-specific coverage for names and speech.

Language coverage by discovery task
TaskLanguagesAssessment
Document OCRAssamese, Bengali, Gujarati, Hindi, Kannada, Malayalam, Marathi, Nepali, Punjabi, Sindhi, Tamil, Telugu and UrduEvaluated on synthetic bilingual ID cards. Validate with your document layouts and scan quality.
Document OCR with partial coverageOdia and SanskritEvaluated on synthetic cards. Recognition needs further improvement and source review.
Identifier and number detectionAll 22 scheduled Indian languages, plus EnglishDigit normalization and number-word lexicons evaluated on synthetic text. Coverage depends on the identifier and extraction quality.
Additional script readingHo, Santali and SoraCharacter recognition evaluated on synthetic text. Personal data detection in these languages needs separate validation.
Names and speechCoverage depends on the configured language modelsAssess names and audio separately for your language mix. Indian-language speech requires evaluation on representative recordings.

Bilingual forms

Inspect records containing English and regional-script fields, then review findings against the original page.

Local numerals

Normalize digits from different scripts before applying identifier patterns and checksum checks.

Your environment

Package the required language models and dependencies locally for the agreed deployment.

Bring representative scans and recordings to your pilot so we can assess language, layout and data-category coverage together.

ILLUSTRATIVE WORKFLOW / SYNTHETIC EXAMPLE

From a source file to a reviewable finding.

A reviewer checks the original field and the normalized digits before confirming the finding.

PIIE is in development. File coverage, detection categories and deployment requirements are agreed during pilot assessment.

PIIE / FINDING CONCEPTSYNTHETIC
Source
bilingual-onboarding.pdf
Category
Phone number candidate
Location
Page 2 · regional-language contact field
Confidence
Illustrative score · requires validation
Review status
Awaiting human review
AFTER DISCOVERY

Give the next team the context it needs.

PIIE finds and locates

Review candidate personal data with file locations, categories and confidence.

PERM coordinates

Use reviewed findings to inform catalogues and privacy workflows through an agreed integration.

Explore PERM ↗

DART redacts

Pass reviewed locations into a scoped redaction workflow for supported content.

Explore DART ↗
START WITH YOUR CONTEXT

Walk through your use case.

Discuss your file formats, data categories and deployment requirements.

inquiry@tekvo.io
Book a 30-minute meeting ↗
EXPLORE MORE

Other ways to use discovery.