VARTEQ

FINTECH CASE STUDY

Mortgage document processing software case study

VARTEQ built machine-learning document processing for a US mortgage technology company. The service reads credit application documents, including handwritten and faxed pages, identifies forms and extracts data for lending workflows.

TRANSFORM WITH VARIOUS TECHNOLOGIES

At a glance

Project type
AI document processing
Client
Undisclosed mortgage technology company
Industry
FinTech
Location
United States
Services
AI Development, Custom Software Development
Faster loan processing
Up to 70%
Eligibility decisions, instead of days
Minutes
Web service, then on-site API
2 stages

OVERVIEW

Project overview

Credit management companies in the US have to check a set of documents for every loan application before deciding who is eligible. VARTEQ built machine-learning document processing for an undisclosed mortgage technology company that serves them, first as a web service and then as an on-site RESTful API for banks and credit associations.

The work grew into a loan processing platform: a CRM that tracks each application, models that score eligibility and risk, and audit trails for every step.

WHERE THEY STARTED

The challenge

Loan applications come with bank statements, tax forms, pay stubs and other documents, many in no set format. Some are handwritten or faxed and hard to read. Staff inspected each one and typed the data into databases by hand, compliance reviewers checked the figures again, and any missing document sent the application back. It was slow, and manual entry led to mistakes.

WHAT WE BUILT

The solution

  1. Web service and on-site API

    • A web service to scan credit application documents, extract the data and enter it into the right databases.
    • A RESTful API that banks and credit associations run on site, returning the data as text or database forms.

  2. Form recognition and data extraction

    • Machine learning that recognizes document images and identifies which credit application form each belongs to.
    • NLP that parses bank statements, tax forms and pay stubs and pulls out the financial values.
    • Many documents processed at once by splitting the data into parallel streams.

  3. Loan workflow and scoring

    • A .NET CRM that tracks each loan application from upload to decision.
    • Regression and neural network models for risk scoring, eligibility and payment forecasts.
    • Payment and reserve calculations from credit and income data, plus credit bureau API integrations.

  4. Monitoring and audit trail

    • Automatic monitoring of every interaction with each application.
    • An audit trail and validation checks that flag anomalies, in real time and in scheduled batches.

SERVICES ON THIS PROJECT

TECHNOLOGIES USED

  • .NET
  • Kendo UI
  • jQuery
  • Python
  • TensorFlow
  • scikit-learn
  • Pandas
  • NumPy
  • SciPy
  • Microsoft SQL Server
  • Microsoft Azure

WHAT CHANGED

The outcome

  • Loan processing time cut by up to 70%.
  • Eligibility decisions in minutes instead of days.
  • Less manual data entry and fewer manual reviews.
  • More consistent checks, with an audit trail at every step.
  • Borrowers see live status updates on their applications.

More client stories

All case studies

START THE CONVERSATION

Facing a similar challenge?

Tell us what you're building or fixing. A VARTEQ partner will reply to arrange a first call.