How it works

How Aadhaar masking works,
step by step.

What you see, what the engine does, and what happens to your image afterwards.

01

Masking through the API

How an application sends an image and gets the masked copy back.

API keyQueueSingle-use results
  1. AuthenticateEvery request carries an API key, and the access rules you set are applied to it: key expiry, images per key, requests per client in a time window, file size and resolution caps, and how long uncollected results are kept. Only a one-way hash of each key is stored.
  2. Send the imageThe service checks the file from its real content (type, size, dimensions, not empty or damaged) before it reaches the engine, and answers at once with a request ID.
  3. Queue and processRequests wait in a bounded queue and are processed in order. When the queue is full, the caller is told how long to wait instead of the request failing.
  4. Check the statusThe caller asks for the status of its request and gets its place in line, then the result: masked, no Aadhaar found, or an error with a clear reason.
  5. Download, onceThe masked image is collected once, with the private token of the request that made it, then deleted from the service.
Good to know
  • The free trial runs this same flow, so you can see how your systems would call it and how access rules behave.
02

Inside the masking engine

What the engine does to your image.

Vision modelTwo OCR engines8 of 12 digits
  1. StraightenThe image's orientation and skew are corrected, so sideways and tilted photos can be read.
  2. Find the cardA trained vision model locates the Aadhaar card. A second detector looks for the QR code.
  3. Read the numberTwo text-recognition engines read the card and pick out the 12-digit number, horizontal or vertical. A card that already shows a masked number is recognised.
  4. MaskThe first 8 digits are covered with a solid block and the last 4 are kept. The QR code is covered completely.
  5. Return in the same formatThe masked image is encoded in its original format (JPEG stays JPEG, PNG stays PNG). An image with no Aadhaar number is returned unchanged.
Good to know
  • Automated masking can miss a card in a very blurry, dark or heavily cropped photo, so check the result for critical documents.
03

What happens to your data

Where your image goes, and when it is gone.

No diskAuto-erase
  1. In memory onlyYour image is held in the service's memory while it waits and while it is masked. It is never written to disk or a database.
  2. Collected onceThe masked result is released once, to the request that made it. Then it is erased.
  3. Never collected?Uncollected results are erased automatically after the retention time set for your deployment.
  4. A minimal logEach request is logged with its time, source IP address, file type and size, a fingerprint of the file, and whether masking succeeded, for audit and to stop misuse. No image, number, file name or card text is logged.
04

Performance and sizing

What it runs on, and how it is sized for your volume.

CPU onlySized to you
  1. No GPU neededMaskly runs on ordinary CPU servers. It has been tested on CPU-only hardware: an Intel Core i7 machine (4 cores, 8 threads, 32 GB of memory) and a 2-vCPU, 8 GB cloud server.
  2. Built for volumeRequests are queued and processed continuously, so a steady stream of uploads or a whole archive is worked through without anyone waiting on a screen.
  3. Sized to your volumeTell us how many documents you handle a day, or how large your archive is and when it must be done. We size the servers with you.
  4. Proven on your documentsBefore go-live, we run Maskly on your own hardware with your kind of documents and agree the numbers with you.
Good to know
  • Large, high-resolution scans take longer than phone photos, so sizing uses your real mix of documents.
  • Need more volume? Talk to our sales team.
05

Inside DocVault

How the same engine works inside DocVault document management.

Auto-mask foldersWorkflowsBulk jobs
  1. Choose where it appliesAn administrator turns masking on and picks folders. Files uploaded there are masked automatically.
  2. Or run it on demandMask a single file, add a masking step to a workflow, or run a bulk job over files already stored.
  3. Every versionEach stored version of a file is masked, so an older copy can't bring the number back.
  4. Nothing to undoThe masked copy replaces the original, so no unmasked copy is kept.
  5. Clear reportsFiles that were skipped (for example, under legal hold or signed) are listed, with the reason.
06

On your own servers

Running the engine in your own network.

ContainerYour network
  1. A containerThe engine ships as a container image that runs alongside your systems, with no outside calls.
  2. CPU onlyIt runs on ordinary CPUs; no GPU is needed.
  3. SizingPlan on at least 8 GB of memory. We size the servers to your volume with you (see Performance and sizing).
  4. Connect your systemsYour applications send images over an internal HTTP API and get the masked image back.
Good to know
  • Talk to us about licensing and setup.

See it run
on your own documents.

Book a demo with your documents, or try the free trial first.

Book a demoTry it free