Powering fraud detection with AI

PitchPoint Solutions is the leading provider of comprehensive fraud and risk mitigation for mortgage applications.

We helped them design a multi-modal LLM solution for validating and parsing government verification forms.

Problems
Solutions

Detection of application tampering was manual and time-consuming.

Used Cloud Vision, Gemini, and Document AI to detect non-text elements and extract text fields.

Annual template changes required regular model tuning and retraining.

Implemented best practices to fine-tune, re-train, and scale ML models on Google Cloud.

Results

Efficiency
99.7%
reduction in the longest loan review step
Systems implemented
4
Time To Deliver
6
weeks

The Process

Discovery

Build Out

Outcomes

Outcomes

What was built

Validation and tampering ML model to identify authenticity and completeness of forms and ensure they have not been tampered with

OCR parser ML model to extract text information relevant to the application

Architecture diagram and supporting technical documentation

API micro service for easy integration