Loan management software case study

Background

Centaur Property Funds Management is a Sydney-based non-bank lender. They make construction loans to property developers, funded by industry super, family offices and high-net-worth investors.

For years they ran their loan book on an Excel model, developed with a financial modelling software firm. Each loan has drawdowns, interest, line fees, investor transfers and covenants to track, and every investor needs their own balances, forecasts and return calculations. As the book grew, the spreadsheet struggled with the volume of data and calculations. Access to the model in their Microsoft environment was also a challenge.

An extensive search for a replacement found nothing off the shelf that met their specific requirements. They decided the best option was to build their own loan management system, with additional tracking and reporting, and far better access for deal teams to view loan positions and covenants.

The work

We built a bespoke web application using React and Python, hosted on AWS.

It includes loan management, investor and fund tracking, forecasting, reporting and more. A daily calculation engine works out balances, interest, line fees and investor returns.

The application has user permissions and a full audit trail and change log. For security:

• Staff sign in with their Microsoft work accounts.
• A web application firewall screens traffic.
• Logs go to the lender's own security monitoring.
• Data is stored in Australia.

The lender owns their code and can download a full backup of their data at any time.

AI and Claude

The lender's team already used Claude, so we brought AI to the loan book in three ways:

1. Ask: an in-app assistant. Staff ask questions about loans and investors in plain English and get answers from live data, with a chart when a trend is easier to see. It runs Claude through Amazon Bedrock in Australia.
2. A Claude connector: we built a custom MCP server that adds the platform to Claude. Staff can ask "Which current loans have the highest LVR?" in claude.ai or Claude Desktop, and Claude fetches the answer. It can also generate almost every report on the platform.
3. A plain-English changelog: AI writes a changelog entry for every platform change we, the developers, make. If a change moves figures on our test loans, the entry lists them.

The connector signs in as the person using it, with their own Microsoft login. Claude only sees what that person can see, and it can't change anything. We worked through the security with the lender's IT advisers before building it.

Services

We created a front-end prototype and worked closely with the client to refine it. We then built the bespoke software in close collaboration with their team. Once it was built, we added the AI features.

The services provided include frontend development, backend development, AWS infrastructure, product design, MCP server development and AI development.

Testimonial