Finnovation, lending · 2019 to 2021

Loan files arrived as scanned IDs and had to be re-keyed by hand

Led by Ansuj Joshi as data scientist at Finnovation Technology, Dec 2019 to Feb 2021

Situation

KYC documents (PAN, Aadhaar) and application text came in as images and free text, and loan officers typed it all into the system.

What we built

OCR REST APIs that extract structured KYC fields and NLP pipelines that score applications from their text.

What changed

Manual loan review fell by 30%.

The problem

The intake step of a loan is reading documents. It is slow and error-prone, and every re-typed field is a chance for a mistake.

What we built

  • OCR services that return structured fields (name, number, date of birth, address) from PAN and Aadhaar images.
  • NLP text-analytics pipelines that feed the credit-scoring model from application text.
  • REST APIs so the existing loan system called them without a rewrite.

How it works

  • Image in, structured JSON out, served over REST so the existing loan system could call it.

What changed

30%

less manual loan review

Source: Ansuj Joshi resume, 2026

Stack

PythonOCRNLPREST

If this looks like your situation

Tell us where it hurts. We will say whether it is worth automating, and what we would build first.