Product 02 · Document OCR

Reads what your customers actually carry.

NICs, driving licences, utility bills, statements and PDFs in Sinhala, Tamil and English. Document OCR returns structured fields with a confidence for each, and refuses anything it cannot read well enough to trust.

nic_front.jpgREAD · 0.98
NIC front
(customer scan)
Full nameK. A. Nimal Perera
NIC number198934501234
Date of birth1989-12-10
Address line 2Officer to enter

0.71: below the 0.95 gate, so not presented as fact.

nic_back_v1.jpg rejected: glare across the bottom-right corner.Replacement requested

Documents

Trained on the paperwork of Sri Lankan lending.

National Identity Card

Old and new formats, Sinhala or Tamil, front and back. Both sides in one scan accepted.

Driving licence

The alternative identity document when a Sinhala or Tamil NIC translation cannot be confirmed.

Utility bills

Electricity, water and telecom bills. Issue date checked against the 3-month rule; name and address compared with the ID.

Bank statements

PDF exports and scanned pages, normalised across local bank layouts for the Statement Analyst.

Salary slips & employer letters

Employer, designation, gross and net pay for the loan file.

Your document

Business registrations, leasing agreements, collateral papers. Tell us what your officers see most.

Quality gates

We reject more than we accept. On purpose.

A wrong value in a core banking record costs far more than a second photo. Every rejection names its reason so the officer knows exactly what to ask for.

Blur“Text is not sharp enough to read. Hold the phone still and retake.”
Glare“Reflection covers the number. Move away from the light.”
Shadow“A shadow hides the left edge.”
Missing corners“The card is cut off. All four corners must be visible.”
Unreadable PDF“This PDF has no readable pages. Please re-export or scan again.”
Wrong document“This looks like a bank statement. This step needs a utility bill.”
Too old“Issued 4 months ago. The limit is 3.”

Confidence gating

A number is either certain, or it is typed by a person.

1

Read

The document is located, de-skewed and read field by field, in its own script.

2

Score

Each field gets its own confidence. Format checks (NIC length, date validity) add or remove trust.

3

Translate

Sinhala and Tamil values are translated to English. The translation is shown next to the original.

4

Present or defer

Only very-high-confidence values are proposed. Anything below the gate is left for the officer to enter.

Where it runs

Inside every Finline analyst, or in your own workflow.

Finline KYC uses it for every identity and address step.
Bank Statement Analyst uses it to read scanned statements and salary slips.
Your systems: standalone access for your own document workflows, on request.

Works with

Send us your worst scans.

The creased NIC, the bill photographed under a tube light. We’ll show you what gets read, what gets rejected, and why.