Document OCR API for Indian lending

Read KYC and loan documents into checked, ready-to-use data

Send a photo or PDF of a PAN, Aadhaar, bank statement, cheque, bill, ITR or salary slip. Get back structured JSON in under a second, every field validated, and a clear flag when a person should take a look.

  • 14 Indian document types
  • Checksums and cross-checks
  • Documents not stored
  • Your cloud or ours
Sample PAN card (synthetic specimen)
panno review needed148 ms
pan
ABCPK1234F
name
RAVI KUMAR
fathers name
SURESH KUMAR
dob
1990-04-12

Real output

What you get back

These are actual responses from the LendOCR API for sample documents. Pick one to see the fields, confidence and checks.

Sample PAN card (synthetic specimen)
Synthetic specimen. No real person's data.
panno review needed148 ms
FieldValueConfidence
panABCPK1234F100%
nameRAVI KUMAR99%
fathers nameSURESH KUMAR100%
dob1990-04-12100%
  • dob valid
  • pan format
  • name matches pan
Raw JSON response
{
  "document_type": "pan",
  "fields": {
    "pan": {
      "value": "ABCPK1234F",
      "confidence": 1
    },
    "name": {
      "value": "RAVI KUMAR",
      "confidence": 0.992
    },
    "fathers_name": {
      "value": "SURESH KUMAR",
      "confidence": 0.998
    },
    "dob": {
      "value": "1990-04-12",
      "confidence": 1
    }
  },
  "checks": {
    "dob_valid": true,
    "pan_format": true,
    "name_matches_pan": true
  },
  "needs_review": false,
  "warnings": [],
  "pages": 1,
  "processing_ms": 148
}

How it works

Three steps, one API call

  1. Send the document

    Upload a photo, scan or PDF from your app or loan system. Password-protected PDFs are fine.

  2. We read and check it

    OCR reads every field, then validators check numbers, dates and totals. Unclear images get a second, more careful read.

  3. You get clean JSON

    Fields with confidence scores, check results and a needs_review flag your workflow can act on.

Why LendOCR

Built for lending, not just reading text

Validation, not only OCR

Aadhaar Verhoeff and GSTIN checksums, passport MRZ check digits, PAN, DL and vehicle formats, salary maths and bank statement balances.

Fewer manual checks

Each document says whether it needs review and why. Clear documents flow straight through.

Bank statement analysis

Salary, EMIs, cheque bounces, cash deposits, average balance, and rows that do not add up.

Private by design

No document storage on the extract API, masked Aadhaar, encrypted background jobs, hashed API keys.

Fast and scalable

About 0.2 s per card on GPU, batch uploads, and background jobs with signed webhooks.

Runs where you need it

Use our API, or run the same containers in your own data centre when data must stay in-house.

For developers

One request

Send the file with your API key. The document type is detected automatically, or you can name it. Errors are standard HTTP with clear messages.

Read the developer guide →

curl https://api.lendocr.com/v1/documents:extract \
  -H "Authorization: Bearer lo_test_..." \
  -F file=@pan.jpg

{
  "document_type": "pan",
  "fields": {
    "pan":  { "value": "ABCPK1234F", "confidence": 1.0 },
    "name": { "value": "RAVI KUMAR", "confidence": 0.99 },
    "dob":  { "value": "1990-04-12", "confidence": 1.0 }
  },
  "checks": { "pan_format": true, "name_matches_pan": true },
  "needs_review": false
}

Accuracy

Measured, not promised

We test every document type on clean, blurred, rotated, compressed, low-resolution, dim and phone-photo copies. On our latest test set of synthetic documents (552 fields across 107 runs), LendOCR read 100% of fields correctly with the automatic engine, with no wrong field left unflagged.

Synthetic test documents are not the real world. During a pilot we measure accuracy on your own documents before you go live.

Questions

Which documents can LendOCR read?

PAN, Aadhaar (front and back), voter ID, driving licence, passport, cheques, bank statements, electricity and water bills, bike and car insurance, ITR-V, salary slips, GST certificates and Udyam certificates. Photos, scans and PDFs, including password-protected PDFs.

Do you store our customers' documents?

No. Documents sent to the extract API are processed in memory and dropped. For background jobs, the file is kept encrypted only until it is processed, then deleted; results are kept encrypted for 24 hours so you can collect them, then erased. Aadhaar numbers are always returned masked.

Does LendOCR prove a document is genuine?

It checks that every number could be real (Aadhaar and GSTIN checksums, passport MRZ check digits, PAN and DL formats, bank statement balances) and flags anything doubtful for review. Proving a document belongs to a person needs a government source such as Income Tax or UIDAI verification, which you use alongside LendOCR.

How fast is it?

A card such as PAN or Aadhaar takes about 0.2 seconds on a GPU; digital PDFs take milliseconds. Large files and bulk uploads can run as background jobs that call your webhook when done.

What does needs_review mean?

Every response says whether a person should check the document: when a required field is missing, the model was unsure, or a check failed. Clear documents go straight through; only doubtful ones reach your team.

Can we run it on our own servers?

Yes. LendOCR ships as Docker containers, so banks and NBFCs that must keep data in-house can run it in their own data centre or cloud account.

How do we start?

Ask for a free test key. It works on every endpoint so your developers can integrate, and test calls are never billed. When you go live, you get a live key with your agreed volume.

Try it on your documents

Get a free test key and integrate in an afternoon.