Pranav Varshney

United Wholesale Mortgage / Engineering case study

Cleaner data for migration and loan classification.

Database normalization for a faster GCP migration, alongside a random forest model for classifying loan outcomes.

Team
Data Engineering
Role
Software Engineering Intern
When
May – August 2024

The work, at a glance

As a software engineering intern in Data Engineering at United Wholesale Mortgage, I worked on two projects: preparing relational data for a cloud migration and using machine learning to classify loan outcomes.

I.

Database normalization

Third normal form (3NF) to reduce duplicate data and accelerate migration to GCP.

II.

Loan classification

A random forest model to classify loan proposals into acceptance or rejection outcomes.

Reduction in duplicate data
45%
Faster GCP migration timeline
3 weeks
Savings from loan classification
$20K

01 / Data engineering

Reduce duplicate data before migration.

I normalized databases using third normal form (3NF), reducing duplicates by 45%. The schema work helped accelerate the migration to Google Cloud Platform (GCP) by three weeks.

The focus was on the structure of the data being migrated. Normalization reduced redundancy in the source databases and shortened the migration timeline.

01 / Project workflow

From relational data to GCP

Full size
Existing databases are normalized into 3NF, reducing duplicates by 45 percent and accelerating the GCP migration timeline by three weeks.
Conceptual sequence of the normalization and migration work. The results connect a change in data quality with a shorter migration timeline.

02 / Machine learning

Classify loan outcomes with a random forest.

I implemented a random forest model to classify loan outcomes, identifying whether a proposal should be accepted or rejected. The classification project saved $20,000.

02 / Project workflow

Classifying loan outcomes

Full size
A random forest model takes loan proposal data and predicts an acceptance or rejection outcome.
Conceptual view of the classification task: loan proposal inputs, a random forest model, and the predicted outcome.

03 / Impact

Measure data quality, time, and cost.

The normalization project reduced duplicate data by 45% and accelerated the GCP migration timeline by three weeks. The loan-classification project delivered $20K in savings.