Aadhaar Masking for KYC Documents [IN]

Stop storing readable Aadhaar numbers Lenders, banks, NBFCs and fintechs collect Aadhaar cards in every loan file. Aadhaar Masking finds the Aadhaar numbers and Aadhaar QR codes in those documents and covers them automatically, in the familiar masked-Aadhaar style (XXXX XXXX 1234), before the files are stored or shared. It runs entirely inside your own AWS account.

Accuracy you can measure

• Tested on 20,000+ real KYC and loan-file documents: card scans, phone photos, e-Aadhaar printouts, UIDAI verification screenshots and multi-page loan PDFs.
• 15,500 of them carried an Aadhaar number or QR code; 15,499 were masked (99.99%). The one miss was a number no person could read either.
• Every release is rerun against labelled test sets before it ships.

Fast and scalable

• Long PDFs are split into chunks that are masked at the same time, each on its own AWS Lambda environment: with 40 running at once, a 500-page PDF comes back in about a minute.
• Images and short PDFs are answered in the same API call, usually in seconds.
• The engine scales out with your load: in our load test, 1,000 files sent up to 100 at a time were all masked, with no errors.
• Serverless throughout - AWS Lambda, Amazon ECS on AWS Fargate and AWS Batch - so there are no servers to size or patch. One stack parameter sets how many environments run at once: 40 by default; raise it for more simultaneous users.

What it masks

• The 12-digit Aadhaar number: the first eight digits covered and the last four left visible (nine covered where the number sits inside a line of text, and all twelve when its reading direction cannot be established - too small, too faint or printed sideways).
• The Aadhaar QR code, fully.
• Numbers on UIDAI verification screenshots, and 12-digit Aadhaar numbers in Excel cells.
• Names, photos, dates of birth and addresses are left as they are, so the document stays usable for KYC.

How it works A detection model finds candidate numbers and QR codes on every page, in every orientation. OCR then reads each candidate and checks it for the marks of an Aadhaar number - twelve digits, the 4-4-4 print, a valid checksum - which keeps PAN numbers, account numbers and dates from being covered.

What it accepts JPG, PNG and TIFF images; PDFs; ZIP archives up to 500 MB, returned as one masked archive with a per-file report; Excel (.xls, .xlsx); and files in your own S3, read and written back by link. Password-protected PDFs are rejected, not masked.

Private by design One CloudFormation template creates the whole service in your AWS account in Asia Pacific (Mumbai): an API behind an Application Load Balancer with AWS WAF (internal to your VPC, or public with your own certificate), the masking engine on AWS Lambda, AWS Batch for archives, an S3 bucket and a DynamoDB table. No document is sent to HabileLabs or any third party. Uploads and masked copies are deleted from your bucket automatically, typically within 24-48 hours. You pay AWS for these resources directly.

Simple pricing Annual contracts of page packs - Starter (2.5 lakh pages a year), Growth (10 lakh), Scale (30 lakh) and Enterprise (1 crore) - combined as needed and added to during the contract, for 12, 24 or 36 months. Masking never stops when you pass your pages: GET /usage and the response headers report the count, and alarms fire at 80% and 100%. Your installs and pages also show in the HabileLabs Marketplace portal, from a daily page count the stack sends (never document content; one setting turns it off). Need a different volume or term? Ask us for a private offer.

Know before you buy Masking is automatic. Handwritten numbers on forms and Virtual IDs (VID) are not masked, and a number no person can read may be missed. To fail safe, a likely number that cannot be read is covered whole, so other print is occasionally covered too. For high-stakes uses, spot-check results.